Skip to content

refactor(workflow)!: remove enabled/disabled agent gate from workflow - #7864

Draft
rh-hemartin wants to merge 1 commit into
mainfrom
remove-enabled-gate-from-workflow
Draft

rh-hemartin wants to merge 1 commit into
mainfrom
remove-enabled-gate-from-workflow

Conversation

@rh-hemartin

@rh-hemartin rh-hemartin commented Sep 30, 2026 •

Copy link
Copy Markdown
Member

Summary

This PR removes the workflow-level gate that used config's agent enabled key to block execution. The same check already exists in the CLI, so the workflow gate was duplication. fullsend run now performs the skip itself: for an agent with enabled: false it exits 0 and relays skipped=true, disabled=true, and the skip reason as step outputs.

It is considered a breaking change because disabled agents now start a workflow and skip at the CLI instead of being blocked up front, and disabled is now a reserved pre-script output key.

Related Issue

Closes #7857, Related to #7608

Changes

  • Remove the enabled check and gate from reusable-dispatch.yml (all stages except fix)
  • Keep a fix-only enabled check in the Route job so a disabled fix agent does not hard-fail the fork-PR and PR-head checks; gate Resolve PR head on it
  • Return the typed agentDisabledError from resolveHarnessForLock so the disabled-agent error contract matches run.go
  • Keep the disabled-agent skip signal in the CLI (disabledSkipResult) rather than the pre-script protocol package
  • Reserve the disabled pre-script output key (hard error if a pre-script emits it)
  • Add behaviour-test steps for disabled agents and status-comment assertions
  • Docs: fix-agent exception, roles workaround limits, error-semantics table row and migration note for the reserved disabled key, and the skipped=true consumer state for disabled agents

Testing

  • make lint passes (stage changes first, then run)
  • Tests added/updated for new or modified logic

Checklist

  • PR title follows Conventional Commits (correct type, ! for breaking changes)
  • Commits are signed off (DCO) — human and human-directed agent sessions only
  • I wrote this contribution myself and can explain all changes in it

@qodo-code-review

Copy link
Copy Markdown

PR Summary by Qodo

Remove redundant agent gating from reusable dispatch workflow

🐞 Bug fix ⚙️ Configuration changes 🕐 10-20 Minutes

Grey Divider

AI Description

• Remove the workflow’s agent-enabled check so dispatch no longer duplicates CLI gating.
• Remove workflow-level agent configuration validation; retain role, PR, and kill-switch checks.
Diagram

graph TD
  A["Incoming event"] --> B["Route stage"] --> C{"Role and PR checks"} --> D["Stage job"] --> E["CLI agent handling"]
  C -->|skipped| F["No stage"]
  subgraph Legend
    direction LR
    _step["Workflow step"] ~~~ _check{"Gate"}
  end
Loading
High-Level Assessment

Removing the duplicate workflow gate is appropriate if the CLI is the intended authority for agent enablement. Retaining the separate role, PR, and kill-switch checks preserves their dispatch-time protections.

Files changed (1) +2 / -70

Bug fix (1) +2 / -70
reusable-dispatch.ymlRemove agent validation and dispatch-time enablement gate +2/-70

Remove agent validation and dispatch-time enablement gate

• Removes workflow-level agents-list validation and the step that skips disabled agents. Route output and PR-head resolution no longer depend on the removed agent-check result; role and PR checks remain.

.github/workflows/reusable-dispatch.yml

@fullsend-ai-review

fullsend-ai-review Bot commented Sep 30, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:54 AM UTC · Completed 12:24 PM UTC

Commit: 68a4973 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5 · Effort: high · Cost: $9.57

@qodo-code-review

qodo-code-review Bot commented Sep 30, 2026 •

Copy link
Copy Markdown

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

Grey Divider


Action required

1. Disabled retro fails agents workflows ✓ Resolved
Description
The stage output in reusable-dispatch.yml no longer checks agent-check.outputs.skipped, so it
routes agents marked enabled: false despite the per-org dispatcher retaining that gate. When a
PR-close event or authorized /fs-retro command routes the disabled retro agent, its stage job
starts and mints a token before fullsend run rejects the agent, rather than skipping the stage.
Code

.github/workflows/reusable-dispatch.yml[113]

+      stage: ${{ steps.role-check.outputs.skipped != 'true' && steps.pr-check.outputs.skipped != 'true' && steps.route.outputs.stage || '' }}
Relevance

●●● Strong

Accepted workflow-gating findings address stage-routing regressions; this exposes disabled agents
before CLI validation and token minting.

PR-#2781
PR-#6455

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
The agents repository invokes the changed reusable workflow and retains the retro role while setting
its agent to enabled: false. The stage expression omits the agent-enabled check, and the remaining
role check reads only roles, so the routed stage can start and mint a token before the action
calls fullsend run; its source resolver then rejects the disabled agent. The scaffolded per-org
workflow still requires the agent check before dispatch, showing the routing difference that the
changed site does not explain.

fullsend -> agents
Rule 1062045: Keep jq payload, stage routing, and secret threading logic in dispatch workflows in sync
.github/workflows/reusable-dispatch.yml[110-114]
.github/workflows/reusable-dispatch.yml[487-512]
.github/workflows/reusable-dispatch.yml[1250-1253]
internal/cli/run.go[5888-5900]
.github/workflows/reusable-dispatch.yml[611-616]
.github/workflows/reusable-dispatch.yml[636-658]
action.yml[442-459]
internal/cli/run.go[5882-5897]
.github/workflows/reusable-dispatch.yml[112-113]
internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml[494-518]
internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml[553-557]
External repo: fullsend-ai/agents, .github/workflows/fullsend.yaml [44-58]
External repo: fullsend-ai/agents, .fullsend/config.yaml [12-19]
External repo: fullsend-ai/agents, .fullsend/config.yaml [42-49]

Agent prompt
The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The agents repository retains the retro role but explicitly disables its agent. The per-repo workflow now starts that stage and reaches a CLI error instead of skipping it, while the per-org dispatcher still gates on the agent check.

## Fix Focus Areas
- .github/workflows/reusable-dispatch.yml[110-113]
- .github/workflows/reusable-dispatch.yml[487-515]
- /cross_repos/agents/.fullsend/config.yaml[42-49]

## Recommended Fix
Restore the enabled-agent check before publishing the routed stage, or add an equivalent early gate that leaves the stage empty for an explicitly disabled agent without removing installed roles. Keep CLI validation as a second line of defense. If the per-repo workflow is intentionally meant to delegate this check to the CLI instead, add an inline comment beside the stage output explaining its divergence from the per-org dispatcher.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools


Grey Divider

Context sources
✅ Compliance rules (platform): 72 rules
✅ Cross-repo context — repo relationships
  Explored: repo: fullsend-ai/experiments (sha: 3f1614fa) — View relationship
  Explored: repo: fullsend-ai/.fullsend (sha: 7c163cad) — View relationship
  Explored: repo: fullsend-ai/pi-xai-vertex (sha: 8f50bdda) — View relationship
  Explored: repo: fullsend-ai/scribe (sha: 0f3eb528) — View relationship
  Explored: repo: fullsend-ai/agents (sha: 1a94080c) — View relationship
  Explored: repo: fullsend-ai/mlflow-assessments (sha: 46f42987) — View relationship
  Explored: repo: fullsend-ai/pi-anthropic-vertex (sha: 960eff35) — View relationship
  Explored: repo: fullsend-ai/metrics (sha: b66b1655) — View relationship
Review mode: ⚖️ Balanced: This removes workflow validation and an execution gate, changing CI behavior and agent configuration semantics, so it warrants a careful single-pass review.

Grey Divider

Tip of the day
💡 Did you know, you can tweak Display settings with a live preview to see your comment before it ships

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

Grey Divider

Qodo Logo

Comment thread .github/workflows/reusable-dispatch.yml Resolved
@codecov

codecov Bot commented Sep 30, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

@fullsend-ai-review

fullsend-ai-review Bot commented Sep 30, 2026 •

Copy link
Copy Markdown

Review

Findings

Medium

  • [protected-path] .github/workflows/reusable-dispatch.yml — This PR modifies a protected path (.github/). It links Enabled gate at workflow level can be dropped #7857 and explains the rationale, but human approval is always required for protected-path changes, regardless of context.

Low

  • [logic-error] .github/workflows/reusable-dispatch.yml:930 — In the new fix-check step, yq stderr is sent to /dev/null, yet the failure handler says "see logs". Nothing diagnostic reaches the logs, and the handler exits 0, so the disabled-fix guard is skipped. fullsend run still skips a disabled agent later, so impact is limited to lost diagnostics.
    Remediation: Capture yq stderr (as the base step did) and print it in the failure handler.
  • [doc-style] docs/cli/run.md:350 — The new ## Early termination and skips section's intro line and both bullets are single long lines. docs/contributing/documentation.md says new prose is written wrapped at about 80 columns, with list continuation lines indented.
    Remediation: Wrap the intro and both bullets at about 80 columns.
  • [backward-incompatible] internal/prescript/prescript.go:191 — disabled is now a reserved pre-script output key, so a pre-script that already emits it hard-fails. This tightens the v1 protocol without a version bump. The PR documents it with a migration note and the ! marker.
    Remediation: Keep the migration note; optionally call out the reserved key in release notes.
  • [backward-incompatible] internal/cli/run.go:669 — fullsend run for a disabled agent now exits 0 and relays skipped=true, disabled=true and reason, where it used to fail. Disabled non-fix agents now start a workflow run before the CLI skips. Documented and marked !; consumers must key off skipped.
    Remediation: Ensure downstream consumers check skipped, not only the exit code.

Earlier findings

  • internal/cli/run.go:7298, api-contract — resolved: resolveHarnessForLock now returns the same typed agentDisabledError (internal/cli/lock.go:656).
  • docs/normative/prescript-output/v1/README.md:201, incorrect-doc — resolved: the consumer-state table now covers the disabled-agent skip.
  • docs/normative/prescript-output/v1/README.md:168, incomplete-doc — resolved: the reserved disabled key is in the hard-errors table and has a migration note.
  • docs/guides/user/customizing-agents.md:517, incorrect-doc — resolved: the Disabling agents section now describes the CLI skip, the fix exception, and the roles limits.
  • internal/prescript/prescript.go:118, pattern-inconsistency — resolved: the PR no longer changes Prepare; skip synthesis lives in disabledSkipResult in run.go.
  • internal/prescript/prescript.go:74, misplaced-abstraction — reclassified low → info: the coupling is deliberate, because reserving the key stops a pre-script from forging the signal action.yml uses to skip reconcile-status.

8 finding(s) below the low severity threshold were not posted.

Previous run

<!-- fullsend:review-findings-v2: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 -->

Review

Findings

Medium

  • [api-contract] internal/cli/run.go:7298 — Prior finding still open: the typed disabled-agent error contract is unchanged since the prior review, and its resolution could not be confirmed. internal/cli/lock.go resolveHarnessForLock still returns the old plain "explicitly disabled" error for the same condition.
    Remediation: Confirm whether the typed error should cover lock.go too; resolve if already addressed.
  • [incorrect-doc] docs/normative/prescript-output/v1/README.md:201 — The consumer state table still says skipped == 'true' means "The pre-script requested a skip", but disabled agents now also relay skipped=true with disabled=true.
    Remediation: Update the true row to cover disabled agents and point consumers at the disabled output.
  • [incorrect-doc] docs/guides/user/customizing-agents.md:517 — Says the dispatch workflow no longer checks whether an agent is enabled, but fix-check in reusable-dispatch.yml still gates fix before setup steps run.
    Remediation: Document fix as the exception to the CLI-level skip.
  • [protected-path] .github/workflows/reusable-dispatch.yml — This PR modifies a protected path. It links Enabled gate at workflow level can be dropped #7857 and explains the rationale, but human approval is always required for protected-path changes.

Low

  • [logic-error] .github/workflows/reusable-dispatch.yml:939 — The "Resolve PR head" step no longer checks steps.fix-check.outputs.skipped. With fix disabled, /fs-fix on a fork PR still fails the fork check (and a PR-fetch failure still fails the route job), defeating the purpose of the retained fix-only gate.
    Remediation: Add steps.fix-check.outputs.skipped != 'true' to the step condition and add a regression test.
  • [pattern-inconsistency] internal/prescript/prescript.go:118 — DisabledResult uses the literal "reason" instead of the reasonKey constant.
    Remediation: Use reasonKey: reason.
  • [misplaced-abstraction] internal/prescript/prescript.go:74 — Disabled-agent skip signalling is added to the pre-script output protocol, though no harness or pre-script runs for a disabled agent.
  • [incomplete-doc] docs/normative/prescript-output/v1/README.md:168 — The Error semantics table omits the new hard error for a pre-script emitting the reserved disabled key.
    Remediation: Add a row for it.
  • [backward-incompatible] internal/prescript/prescript.go:204 — disabled is now a reserved pre-script key that hard-fails, and the spec stays at v1.
    Remediation: Note it in the migration notes.
  • [backward-incompatible] internal/cli/run.go:669 — fullsend run for a disabled agent now exits 0 and relays skip outputs. Marked breaking and documented.

Earlier findings

  • internal/prescript/prescript.go:75 api-shape, internal/prescript/prescript.go:91 doc-style, internal/cli/run.go:7298 error-handling-idiom, action.yml:589 design-smell, pkg/behaviourtest/steps/statuscomment.go:14 scope-creep, .github/workflows/reusable-dispatch.yml:126 permission-expansion, .github/workflows/reusable-dispatch.yml:927 fail-open — reclassified, tracked at info; the same concerns remain and none was raised higher.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR

7 finding(s) below the low severity threshold were not posted.

Previous run (2)

<!-- fullsend:review-findings-v2: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 -->

Review

Findings

Medium

  • [api-contract] internal/cli/run.go:7303 — agentDisabledError.Error() now returns agent "<name>" is disabled in config. It used to return ... is explicitly disabled in config. internal/cli/run_test.go is not changed by this PR, and TestResolveAgentSource_* (lines 1487, 1510, 1532) still assert "explicitly disabled". Those tests fail with the new wording.
    Remediation: Keep the old wording in Error() and use SkipReason() only for the skip message, or update the three tests to use errors.As on *agentDisabledError.
  • [protected-path] .github/workflows/reusable-dispatch.yml — The PR modifies a protected path. It links Enabled gate at workflow level can be dropped #7857 and explains the rationale. Human approval is always required for protected-path changes, regardless of context.

Low

  • [logic-error] .github/workflows/reusable-dispatch.yml:939 — fix-check sets skipped=true, but only the job-level stage output reads it. The Resolve PR head for issue_comment events condition no longer tests a skip flag, and steps.route.outputs.stage is still fix. For a disabled fix agent, a fork PR or a PR-fetch failure still hard-fails the Route job. This is the failure the TODO on fix-check says the step prevents.
    Remediation: Add steps.fix-check.outputs.skipped != 'true' to the pr-head step condition.
  • [pattern-inconsistency] internal/prescript/prescript.go:118 — DisabledResult uses the literal "reason" instead of the reasonKey constant, although the entry before it uses disabledKey.
    Remediation: Use reasonKey: reason.
  • [misplaced-abstraction] internal/prescript/prescript.go:74 — The CLI's disabled-agent signal (disabledKey, DisabledResult) and the reserved disabled key in the normative v1 pre-script spec are added to package prescript. That mixes the pre-script output contract with CLI config evaluation, which runs before any pre-script is resolved.
    Remediation: Consider modelling the disabled skip in internal/cli or a dedicated helper.
  • [backward-incompatible] internal/prescript/prescript.go:204 — disabled is now a reserved key, and ParseFile hard-fails if a pre-script emits it (including case variants). The v1 spec was changed in place. External pre-scripts that already emit disabled would start failing. No in-repo pre-script emits it.
    Remediation: Note it in the release notes, or warn for one release before the hard error.
  • [backward-incompatible] internal/cli/run.go:669 — fullsend run now exits 0 for an agent with enabled: false. It used to exit non-zero. Direct callers that relied on the failure will now see success. The title carries ! and the body says the change is breaking.
    Remediation: Keep the breaking-change note in the release notes.

Earlier findings

  • e2e/behaviour/features/triage/triage.feature:30, test-inadequate — resolved: owners.go now unescapes the quoted log text, owners_test.go covers the exact scenario text, and the scenario also checks status-comment absence.
  • internal/prescript/prescript.go:75, api-shape — reclassified to info: nothing outside the package reads Outputs["disabled"].
  • action.yml:589, design-smell — reclassified to info: the duplicate gate is deliberate and each layer has its own test.
  • pkg/behaviourtest/steps/statuscomment.go:14, scope-creep — reclassified to info: the step is small and unit-tested.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR

7 finding(s) below the low severity threshold were not posted.

Previous run (3)

<!-- fullsend:review-findings-v2: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 -->

Review

Findings

Medium

  • [test-inadequate] e2e/behaviour/features/triage/triage.feature:30 — The step the triage workflow logs contain "Run skipped: agent \"triage\" is disabled in config" cannot match its registered Godog expression ^the triage workflow logs contain "([^"]+)"$ (pkg/behaviourtest/steps/owners.go:32). [^"]+ stops at the first embedded quote, and Gherkin does not unescape quotes in step text. The step is therefore undefined, so the disabled-agent scenario never checks the skip log, and the following no-status-comment assertion is not exercised.
    Remediation: Use a quote-free substring such as is disabled in config, or extend the step expression and handler to decode escaped quotes. Add a step-registration or dry-run check covering this scenario.
  • [protected-path] .github/workflows/reusable-dispatch.yml — This PR modifies a protected path. The PR links Enabled gate at workflow level can be dropped #7857 and explains the rationale, but human approval is always required for protected-path changes regardless of context.

Low

  • [logic-error] .github/workflows/reusable-dispatch.yml — Carried PR-level finding on the dispatch workflow; the current diff gives no evidence that it has been resolved.
  • [misplaced-abstraction] internal/prescript/prescript.go:74 — Config-based agent suppression is bound to the pre-script output protocol package. disabledKey, DisabledResult and the reserved disabled key are added to internal/prescript only to reuse prescript.Relay, so the protocol must explicitly forbid pre-scripts from emitting the key.
  • [design-smell] action.yml:581 — The Finalize orphaned status comment step is gated twice: the step-level if: has steps.run.outputs.disabled != 'true' and the script also exits early on WAS_DISABLED. The script check cannot be reached when the step is skipped. It is harmless defence in depth.
  • [scope-creep] pkg/behaviourtest/steps/statuscomment.go:16 — The PR adds and documents both a positive and a negative status-comment step, but only the negative one is used by the new scenario.
  • [api-shape] internal/prescript/prescript.go:75 — Carried API-shape finding on the new disabledKey constant and DisabledResult constructor; unchanged and still present.
  • [backward-incompatible] internal/prescript/prescript.go:204 — The disabled output key is now reserved, case-insensitively. A pre-script that already emits it now fails the run. The PR is marked breaking and documents this in the normative spec.
    Remediation: Mention the reservation in the release notes.
  • [backward-incompatible] internal/cli/run.go:641 — fullsend run now exits 0 and relays skipped=true, disabled=true and reason for a disabled agent, where it used to return an error. Callers that treated a non-zero exit as the disabled signal must now check the outputs.
    Remediation: Make sure the release notes and docs/cli/run.md tell consumers to check the disabled and skipped outputs.
  • [sub-agent-failure] N/A — The style-conventions sub-agent did not return findings: it timed out after 900s. Its prior findings stay open and unverified by that dimension.

Earlier findings

  • pkg/behaviourtest/steps/dispatch.go:275, logic-error — resolved: givenDisabledAgent now removes every case-insensitive matching entry before appending the disabled one, and TestGivenDisabledAgent_ReplacesMultipleEntries covers both orderings.
  • docs/guides/dev/behaviour-testing.md:444, incomplete-doc — resolved: the new disabled-agent and status-comment steps are now documented.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (4)

<!-- fullsend:review-findings-v2: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 -->

Review

Findings

Medium

  • [test-inadequate] e2e/behaviour/features/triage/triage.feature:30 — The new scenario does not prove that execution reaches the CLI skip. The removed workflow gate also let the dispatch complete successfully with no agent run and no status comment, and it emitted a notice containing "is disabled in config". If that gate came back, every assertion would still pass, so the scenario's stated contract is not pinned.
    Remediation: Assert that the triage agent job and its Run fullsend step actually execute, and match the CLI-specific message (Run skipped: agent "triage" is disabled in config) or the disabled=true output rather than the shared substring.
  • [protected-path] .github/workflows/reusable-dispatch.yml — This PR modifies a protected path (removes the Validate agents config and Check agent is enabled steps and the agent-check references in the stage output and the pr-head step condition). The PR links Enabled gate at workflow level can be dropped #7857 and explains the rationale, but human approval is always required for protected-path changes regardless of context.

Low

  • [incomplete-doc] docs/guides/dev/behaviour-testing.md:444 — The PR adds the the (issue|pull request) has / does not have a status (comment|marker) and status notification comments are disabled steps but documents only the disabled agent "<name>".
    Remediation: Add short entries for the status-comment steps beside the the disabled agent entry.
  • [logic-error] .github/workflows/reusable-dispatch.yml — Carried forward from the prior review; nothing in the diff shows it resolved. Disabled agents now enter the stage job and its setup steps before the CLI skips them.
  • [logic-error] pkg/behaviourtest/steps/dispatch.go:275 — givenDisabledAgent replaces only the first matching entry. For a disabled-then-enabled pair the final enabled entry is left alone; for an enabled-then-disabled pair it produces two disabled entries, which fails duplicate validation.
    Remediation: Collapse every entry matching the agent name into a single disabled entry, keep unrelated entries, and add tests for both pair orders.
  • [misplaced-abstraction] internal/prescript/prescript.go:74 — Agent enablement (disabled=true, DisabledResult) is built into the pre-script output protocol package even though a disabled agent never runs a pre-script.
    Remediation: Relay the disabled-agent skip outputs from internal/cli (or a dedicated helper) instead of adding CLI-only state to the pre-script protocol package.
  • [design-smell] action.yml:581 — steps.run.outputs.disabled is checked in both the step if: and the script, so the in-script WAS_DISABLED early exit can never trigger.
    Remediation: Keep the check in one place.
  • [scope-creep] pkg/behaviourtest/steps/statuscomment.go:16 — The status notification comments are disabled step, ResetStatusCommentNotifications and the matching teardown change status_notifications config, are unrelated to Enabled gate at workflow level can be dropped #7857, and no feature scenario uses them.
    Remediation: Remove the unused step and its helpers; keep only the assertions the disabled-agent scenario needs.
  • [api-shape] internal/prescript/prescript.go:75 — DisabledKey is exported while skippedKey and reasonKey are not, and nothing outside the package uses it. DisabledResult also uses the literal "reason" instead of reasonKey.
    Remediation: Rename to disabledKey and use reasonKey.
  • [backward-incompatible] internal/prescript/prescript.go:95 — disabled is now a reserved pre-script output key, so existing pre-scripts that emit it are rejected. The PR's breaking-change note does not mention the reserved key.
    Remediation: Mention the newly reserved key in the breaking-change or release notes.
  • [backward-incompatible] internal/cli/run.go:641 — Disabled agents now start a workflow and fullsend run exits 0 with skipped=true and disabled=true; consumers see a new disabled output and runs that previously never started. Documented as breaking and marked !.

Earlier findings

  • f_fd88c9ea4befa35d — reclassified (low → info): the reservedKeys comment starts with the identifier as godoc expects; the concern is cosmetic.
  • f_1665d028de46609d — reclassified (low → info): a typed error matched with errors.As is idiomatic and agentName is used.
  • f_be3625d5ad560f6f — reclassified (stays info): the workflow stage output no longer references steps.agent-check; permissions are unchanged.
  • f_9ff98772572485cf — reclassified (stays info): the removed gate failed open on yq errors; the CLI check is now the sole gate and fails closed on explicit disablement.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (5)

<!-- fullsend:review-findings-v2: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 -->

Review

Findings

Medium

  • [protected-path] .github/workflows/reusable-dispatch.yml — The PR modifies a protected governance/infrastructure path (removes the "Validate agents config" and "Check agent is enabled" steps and the agent-check term in the stage output and downstream if: conditions). It links Enabled gate at workflow level can be dropped #7857 and explains the rationale, but human approval is always required for protected-path changes.
  • [test-inadequate] e2e/behaviour/features/triage/triage.feature:30 — The disabled-agent scenario does not show that the Triage job reached fullsend run. The removed workflow gate's ::notice:: also contains the asserted substring "is disabled in config", and ensureTriageWorkflowComplete only checks workflow-level success. The scenario would therefore pass against the pre-PR behavior too.
    Remediation: Assert that the Triage job and its Run fullsend step executed successfully, and match the CLI-specific skip output (e.g. Run skipped: agent "triage" is disabled in config).

Low

  • [incomplete-doc] docs/guides/dev/behaviour-testing.md:442 — New steps in pkg/behaviourtest/steps/statuscomment.go (status notification comments are disabled, the (issue|pull request) has / does not have a status (comment|marker)) are not documented. (Reclassified from medium: the workflow-log step is pre-existing.)
    Remediation: Add reference entries for these step phrases next to the other step definitions.
  • [logic-error] .github/workflows/reusable-dispatch.yml — Dropping the workflow's agents-shape validation also drops duplicate-name rejection for agents inherited from config.base.yaml. CLI validation runs after keyed merging, so duplicate same-state overlay entries now collapse silently (last writer wins).
    Remediation: Validate each layer's local agent entries (LocalAgentEntries()) before merging, and add a regression test.
  • [logic-error] pkg/behaviourtest/steps/dispatch.go:275 — givenDisabledAgent replaces only the first matching entry, so with a valid disable-then-enable pair it returns success without disabling the agent (reverse order yields duplicate disabled entries). Impact is limited to the test helper.
    Remediation: Replace all matching entries with a single disabled entry and test both orderings.
  • [misplaced-abstraction] internal/prescript/prescript.go:75 — DisabledKey/DisabledResult live in the pre-script IPC package and disabled is reserved in the normative pre-script spec, but disabled-agent state is a static config decision made before any pre-script runs.
    Remediation: Decouple the CLI's disabled-agent CI output relay from package prescript.
  • [design-smell] action.yml:581 — The step if: already excludes disabled == 'true', so the inline WAS_DISABLED check in the script is unreachable.
    Remediation: Remove the redundant in-script check.
  • [api-shape] internal/prescript/prescript.go:75 — DisabledKey is exported while skippedKey/reasonKey are not, and it is only used inside the package.
    Remediation: Rename to disabledKey and update in-package uses.
  • [doc-style] internal/prescript/prescript.go:91 — The reservedKeys comment uses bare key names and says keys differing "only by case" are rejected, which is misleading for disabled (rejected unconditionally).
    Remediation: Quote key names and state that disabled is rejected in any case.
  • [scope-creep] pkg/behaviourtest/steps/statuscomment.go:16 — Adds an unused status notification comments are disabled step with cleanup helpers, authorized by neither Enabled gate at workflow level can be dropped #7857 nor ci(workflows): reduce agent workflows to the Fullsend invocation contract #7608 and unused by any scenario.
    Remediation: Remove it or add a scenario that uses it.
  • [error-handling-idiom] internal/cli/run.go:7035 — agentDisabledError.Error() ("explicitly disabled") differs from the skip reason runAgent builds ("is disabled"); the error type does not carry the canonical reason.
    Remediation: Have the error type expose the canonical reason, or use a sentinel with errors.Is.
  • [backward-incompatible] internal/cli/run.go:641 — fullsend run now exits 0 for a disabled agent where it formerly errored. The ! marker, BREAKING CHANGE footer and docs/cli/run.md update are present.
    Remediation: Make release notes tell callers to read the skipped/disabled outputs, not the exit code.
  • [backward-incompatible] internal/prescript/prescript.go:95 — disabled is now a reserved key, so a pre-script emitting it is hard-rejected. No in-repo pre-script does; external harnesses could not be checked.
    Remediation: Check known external harnesses and mention the new reserved key in release notes.

Earlier findings

  • f_57f86c1d7c451639 — reclassified (medium → low): the workflow-log step is pre-existing, so only the new status-comment steps are undocumented.
  • f_be3625d5ad560f6f — reclassified (low → info): early token/secret exposure for disabled agents is justified by Enabled gate at workflow level can be dropped #7857/ci(workflows): reduce agent workflows to the Fullsend invocation contract #7608; permission scopes are unchanged.
  • f_9ff98772572485cf — reclassified (low → info): no CLI fail-open demonstrated; the disabled return precedes harness load, token minting, pre-script and sandbox creation (internal/cli/run.go:641-653).
  • f_8b7d64f72ea79e51 — resolved: not reproduced at the rebased head; the title carries ! and the commit has a BREAKING CHANGE footer.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR
Previous run (6)

<!-- fullsend:review-findings-v2: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 -->

Review

Findings

Medium

  • [test-inadequate] e2e/behaviour/features/triage/triage.feature:30 — The new log assertion has no matching step definition. pkg/behaviourtest/steps/owners.go:32 registers ^the triage workflow logs contain "([^"]+)"$. [^"]+ stops at the first embedded quote and $ then fails, so the escaped quotes around triage never match. The scenario ends with an undefined step and never verifies the disabled-agent message or the following no-status-comment assertion.
    Remediation: Use a quote-free substring supported by the existing step (e.g. is disabled in config), or extend the step definition to accept and unescape embedded quotes. Verify the scenario has no undefined steps.
  • [incomplete-doc] docs/guides/dev/behaviour-testing.md:442 — The new behaviour step (a|the) disabled agent "<name>" [in config] (pkg/behaviourtest/steps/dispatch.go) is not documented under ### Dispatch step reference, which still lists only a disabled custom harness "<name>" with:.
    Remediation: Add an entry for the disabled agent "<name>" saying it sets enabled: false on the agent entry in .fullsend/config.yaml.
  • [protected-path] .github/workflows/reusable-dispatch.yml — This PR modifies a protected path. It links Enabled gate at workflow level can be dropped #7857 and explains the rationale, but human approval is always required for protected-path changes regardless of context.

Low

  • [backward-incompatible] internal/prescript/prescript.go:189 — disabled is now a reserved prescript output key; a pre-script emitting disabled=... now fails with a hard parse error where it was previously an ordinary output. The spec documents it as CLI-emitted only and the PR carries !, so this is a narrow break.
    Remediation: Mention the reserved key in the release notes, or warn-and-ignore for one release.
  • [backward-incompatible] internal/cli/run.go:641 — fullsend run now exits 0 and relays skipped=true and disabled=true for an explicitly disabled agent; before, it failed. Callers that relied on a nonzero exit now see a successful skip, and every disabled default agent now starts a workflow run. The BREAKING CHANGE footer declares this.
  • [misplaced-abstraction] internal/prescript/prescript.go:75 — The disabled-agent skip happens in CLI config resolution before any pre_script runs, yet it is signalled through the pre-script protocol package (disabledKey in reservedKeys, disabled in the pre-script spec, a synthetic prescript.Result in internal/cli/run.go). This couples a CLI lifecycle signal to the pre-script key namespace.
  • [design-smell] action.yml:581 — The Finalize orphaned status comment step is suppressed twice: by the step-level if: and by the WAS_DISABLED early exit in the bash body (line 605). The if: already excludes disabled runs, so the bash early exit is unreachable under Actions.
  • [api-shape] internal/prescript/prescript.go:75 — disabledKey is unexported, so internal/cli/run.go:647 hardcodes "disabled". prescript.Result has typed fields for Skipped and Reason but none for disabled-agent runs, and callers duplicate the reason into Outputs["reason"].
    Remediation: Export the constant or add a helper such as prescript.DisabledResult(reason).
  • [doc-style] internal/prescript/prescript.go:91 — The reservedKeys comment says it lists keys "the protocol itself interprets", but ParseFile interprets only skipped and reason; disabled is reserved only to prevent case collisions and pre-script emission of a CLI-relayed key.
    Remediation: Update the comment to cover both kinds of keys.
  • [fail-open] .github/workflows/reusable-dispatch.yml — Removing the Validate agents config step loses duplicate-entry rejection for layered configs: the CLI validates only the merged AgentEntries(), which collapses overlay entries matching a parent agent into one. Input is trusted maintainer-authored config, so this is a validation-strictness gap, not an access-control bypass.
    Remediation: Optionally validate each layer's raw agent entries before merging and add a regression test.
  • [permission-expansion] .github/workflows/reusable-dispatch.yml:122 — Removing the agent enabled gate from the stage output lets explicitly disabled agents run credential-bearing stage jobs (id-token: write, role-token minting) before fullsend run checks enablement. permissions: blocks are unchanged and the CLI exits before sandbox creation, so no agent code runs with the token. This is the trade-off authorized by Enabled gate at workflow level can be dropped #7857; confirm it is acceptable.
    Remediation: If not acceptable, keep a trusted-config enablement preflight before token minting.

Earlier findings

  • f_305fa287de51a9ce — resolved_by_change: internal/prescript/prescript.go:188-190 now rejects any pre-script-emitted disabled key, and the v1 spec lists it as CLI-emitted only.
  • f_8b7d64f72ea79e51 — reclassified to info: the ! marker and BREAKING CHANGE footer are present.

Next steps:

  • /fs-fix — agent addresses review findings automatically
  • /fs-fix <your instruction> — agent fixes with your specific guidance
  • Push commits directly — review re-runs automatically on push
  • /fs-fix-stop — disable automatic fix runs for this PR

fullsend-ai-review[bot]

This comment was marked as outdated.

@jflowers jflowers left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Council Verdict: REQUEST CHANGES

Reviewers: adversary, architect, fullsend, guard, testing, sre, curator, envoy, herald, scribe · Iterations: 1

1. [HIGH] fix(workflow) type + "not a breaking change" claim rest on an unstated intent, and are unactionable as written. (envoy, herald, sre, guard) — the body asserts "not a breaking change" because disabled agents remain blocked, but the CLI blocks by failing the job (red CI) after the workflow gate previously skipped it (green, zero-cost). Resolve by intent: (a) if skip→fail is unintended, restore skip semantics (the line-113 regression, already covered by bot review); (b) if fail-on-disabled is the intended policy, the change is breaking for consumers pinning reusable-dispatch.yml@<ref> and must carry ! + BREAKING CHANGE: trailer — or be retyped ci(workflow) (it is a CI change, not a user-visible fix). Either way the current "behavior retained, therefore not breaking" framing conflates skip with fail.

2. [MEDIUM] Dispatch-sync contract violated. (architect, testing, sre, fullsend, curator, scribe, herald) — docs/contributing/workflow-contracts.md:3 asserts scaffold dispatch.yml and reusable-dispatch.yml share identical routing logic, but internal/scaffold/fullsend-repo/.github/workflows/dispatch.yml:435,494-521,557 still retains the removed gate/validation. Mirror the removal, or record the intentional divergence (per-org deprecated per ADR 44).

3. [MEDIUM] Stale user doc — docs/guides/user/customizing-agents.md:533 still states enabled: false "prevents the agent from dispatching." (curator, scribe, architect, sre)

4. [MEDIUM] Stale test rationale — internal/config/repo_overlay_test.go:30-43 comment + assertion message name the removed agent-check mechanism. (testing, architect, sre, fullsend, herald)

5. [LOW] ADR 0091 references the removed reusable-workflow guard. (curator, sre)


This review was generated by /uf.review-council (AI-assisted).

@rh-hemartin
rh-hemartin force-pushed the remove-enabled-gate-from-workflow branch from 68a4973 to 9867bb8 Compare October 1, 2026 08:41
@rh-hemartin

Copy link
Copy Markdown
Member Author

@jflowers let me know what you think of the new approach.

@rh-hemartin rh-hemartin changed the title fix(workflow): remove enabled/disabled agent gate from workflow refactorworkflow)!: remove enabled/disabled agent gate from workflow Oct 1, 2026
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 1, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ❌ Failure (resolving persona models: persona "correctness": a Bash(...) allowlist in persona frontmatter is not supported yet and would not restrict the child; declare plain Bash and restrict it on the agent i…) · Started 8:43 AM UTC · Completed 8:44 AM UTC

Commit: 9867bb8 · View workflow run →

Runtime: pi · Model: sonnet · Effort: high

@rh-hemartin

Copy link
Copy Markdown
Member Author

/fs-review

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 1, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:52 AM UTC · Completed 12:33 PM UTC

Commit: 9867bb8 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5 · Effort: high · Cost: $2.97

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 1, 2026 •

Copy link
Copy Markdown

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) anchored to prior review: breaking CI dispatch workflow modifications across 26 files and high git churn are balanced by strong test coverage from an established contributor.

Previous run

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) anchored to prior review: breaking CI dispatch workflow modifications across 24 files and high git churn are balanced by strong test coverage from an established contributor.

Previous run (2)

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) anchored to the prior review: breaking CI dispatch workflow modifications across 23 files and high git churn are balanced by strong test coverage from an established contributor.

Previous run (3)

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) anchored to the prior review: the Tier 1 signal bins are unchanged across 22 files and 849 lines, with breaking CI dispatch workflow modifications and high-churn hotspot files balanced by strong test coverage from an experienced contributor.

Previous run (4)

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) as the PR maintains its risk profile across 24 files and 1,022 lines with CI workflow modifications, breaking dispatch changes, and high-churn hotspot files, balanced by solid test coverage and an experienced contributor.

Previous run (5)

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) as the rebased PR keeps the same risk profile across 24 files and about 1,026 lines with CI workflow modifications and breaking dispatch changes across high-churn files, balanced by strong test coverage and familiar authorship.

Previous run (6)

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Score remains 3 (elevated) as the diff expanded to 23 files and 996 lines with CI workflow modifications and breaking dispatch changes across high-churn files, balanced by strong test coverage and familiar authorship.

Previous run (7)

<!-- fullsend:risk-assessment -->
Risk Assessment: elevated (3/5)

Details

Elevated risk reflects CI workflow and protected path modifications with breaking dispatch behavior changes and high churn on touched files, with the score increasing from prior moderate (2) because the diff expanded from 8 to 16 files (542 lines, large blast radius).

Previous run (8)

<!-- fullsend:risk-assessment -->
Risk Assessment: moderate (2/5)

Details

Moderate risk reflects CI workflow modifications, a protected path change, and high churn on dispatch and runner files, balanced by modest diff size, an established contributor, strong test coverage, and close alignment with the linked issue scope.

Previous run (9)

<!-- fullsend:risk-assessment -->
Risk Assessment: moderate (2/5)

Details

Moderate risk reflects CI workflow modifications, a protected path change, and high churn on dispatch and runner files, balanced by modest diff size, an established contributor, strong test coverage, and close alignment with the linked issue scope.

Previous run (10)

<!-- fullsend:risk-assessment -->
Risk Assessment: moderate (2/5)

Details

Moderate risk preserved at 2 as CI workflow modification, lack of rollback flag, and high churn on dispatch and CLI files are balanced by modest change size (7 files, 186 lines), established contributor, and close alignment with linked issue scope.

Previous run (11)

Risk Assessment: moderate (2/5)

Details

Moderate risk preserved at 2 as CI workflow modification, lack of rollback flag, and high churn on dispatch and CLI files are balanced by modest change size (8 files, 190 lines), established contributor, and close alignment with linked issue scope.

Previous run (12)

Risk Assessment: moderate (2/5)

Details

Moderate risk preserved at 2 as CI workflow modification, lack of rollback flag, and high churn on dispatch and CLI files are balanced by modest change size (8 files, 191 lines), established contributor, and close alignment with linked issue scope.

Previous run (13)

Risk Assessment: moderate (2/5)

Details

Moderate risk preserved at 2 as CI workflow modification, lack of rollback flag, and high churn on dispatch and CLI files are balanced by modest change size (8 files, 281 lines), established contributor, and close alignment with linked issue scope.

Previous run (14)

Risk Assessment: moderate (2/5)

Details

Moderate risk preserved at 2 as CI workflow changes, lack of rollback flag, and high churn on dispatch and CLI files are balanced by small file count (8 files), established contributor, and close alignment with linked issue scope.

Previous run (15)

Risk Assessment: moderate (2/5)

Details

Moderate risk preserved at 2 as CI workflow changes, lack of rollback flag, and high churn on dispatch and CLI files are balanced by small file count (7 files), established contributor, and close alignment with linked issue scope following addition of CLI skip handling.

Previous run (16)

Risk Assessment: moderate (2/5)

Details

Moderate risk due to CI workflow modifications and active multi-author contention on dispatch files, balanced by small file count, an established contributor, and close alignment with linked issue scope.

Previous run (17)

Risk Assessment: moderate (2/5)

Details

Moderate risk due to CI workflow modifications and active multi-author contention on dispatch files, balanced by small file count, an established contributor, and close alignment with linked issue scope.

Previous run (18)

Risk Assessment: elevated (3/5)

Details

Elevated risk due to modification of core CI dispatch workflow without rollback toggles, introducing a breaking change with unaddressed trade-off discussion in the linked issue, combined with high churn and author contention across touched files.

fullsend-ai-review[bot]

This comment was marked as outdated.

@jflowers

jflowers commented Oct 1, 2026

Copy link
Copy Markdown
Contributor

@rh-hemartin thanks for the ping. The new approach reads clean1 The refactor(workflow)! + BREAKING CHANGE: trailer resolves the type/framing point I raised earlier. A few things still seem to apply at this head, mostly confirming intent rather than re-litigating:

  1. Own-repo retro CI. .fullsend/config.yaml still keeps retro in roles with enabled: false, so after merge every PR close and /fs-retro runs the stage job through mint-token/checkout and then fails at fullsend run retro until the CEL migration lands. Is that self-inflicted red CI accepted for now? If there's a tracking issue for the migration, mind dropping a pointer in the body?

  2. User docs. customizing-agents.md now says disabling "prevents executing via fullsend run, which fails," but skips the CI consequence (the stage job runs and fails, rather than skips). Worth one sentence? config-reference.md's enabled bullet is untouched too.

  3. Validate agents config. You already said it's going too. Mind widening Enabled gate at workflow level can be dropped #7857 (or the "Changes" bullet) to name it, so the scope is recorded instead of implicit?

  4. workflow-contracts.md. The new "shouldn't be in sync" sentence sits next to the still-enforced OWNERS-parity rule. One clause clarifying that only the routing/gate logic is relaxed would stop a reader from skipping OWNERS sync.

  5. Title nit. The PR title is still refactorworkflow)!: (missing the () while the commit message is correct. GoReleaser reads the title, so it'd land in "Others" instead of "Refactoring."

Nothing new or blocking here, just confirming the remaining doc/scope gaps are closed or consciously deferred.

@rh-hemartin

Copy link
Copy Markdown
Member Author

Own-repo retro CI. .fullsend/config.yaml still keeps retro in roles with enabled: false, so after merge every PR close and /fs-retro runs the stage job through mint-token/checkout and then fails at fullsend run retro until the CEL migration lands. Is that self-inflicted red CI accepted for now? If there's a tracking issue for the migration, mind dropping a pointer in the body?

Yes, this is accepted for now. Also soon roles won't be gating anyways.

User docs. customizing-agents.md now says disabling "prevents executing via fullsend run, which fails," but skips the CI consequence (the stage job runs and fails, rather than skips). Worth one sentence? config-reference.md's enabled bullet is untouched too.

Changed customizing-agents.md, config-reference.md can stay the way it is.

Validate agents config. You already said it's going too. Mind widening Enabled gate at workflow level can be dropped #7857 (or the "Changes" bullet) to name it, so the scope is recorded instead of implicit?

Added.

workflow-contracts.md. The new "shouldn't be in sync" sentence sits next to the still-enforced OWNERS-parity rule. One clause clarifying that only the routing/gate logic is relaxed would stop a reader from skipping OWNERS sync.

Well, there shouldn't be barely any sync at all, per-org mode will go away soon.

Title nit. The PR title is still refactorworkflow)!: (missing the () while the commit message is correct. GoReleaser reads the title, so it'd land in "Others" instead of "Refactoring."

Fixed.

@rh-hemartin rh-hemartin changed the title refactorworkflow)!: remove enabled/disabled agent gate from workflow refactor(workflow)!: remove enabled/disabled agent gate from workflow Oct 2, 2026
@rh-hemartin
rh-hemartin force-pushed the remove-enabled-gate-from-workflow branch from 9867bb8 to 7f5327c Compare October 2, 2026 06:53
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 2, 2026 •

Copy link
Copy Markdown

🤖 Review · ⚠️ Cancelled · Started 6:55 AM UTC · Ended 7:01 AM UTC

Commit: 7f5327c · View workflow run →

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review sweep: one finding, inline below.

Comment thread .github/workflows/reusable-dispatch.yml Resolved
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 20, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 6:07 AM UTC · Completed 6:24 AM UTC

Commit: c8381ee · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $4.47

@fullsend-ai-review fullsend-ai-review Bot added risk/elevated PR risk: elevated and removed risk/moderate PR risk: moderate labels Oct 20, 2026
fullsend-ai-review[bot]

This comment was marked as outdated.

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 20, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 7:01 AM UTC · Completed 7:15 AM UTC

Commit: 813bb69 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $3.46

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 20, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 8:01 AM UTC · Completed 8:23 AM UTC

Commit: 141a68d · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $4.78

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 20, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 8:55 AM UTC · Completed 9:14 AM UTC

Commit: ab04859 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $4.17

fullsend-ai-review[bot]

This comment was marked as outdated.

@waynesun09 waynesun09 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review findings at ab04859 (inline below).

Comment thread docs/guides/user/customizing-agents.md Resolved
Comment thread .github/workflows/reusable-dispatch.yml Resolved
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 21, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 8:59 AM UTC · Completed 9:22 AM UTC

Commit: 3d536f1 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $4.07

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 30, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 10:46 AM UTC · Completed 11:04 AM UTC

Commit: fd004f1 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $4.27

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 30, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 11:44 AM UTC · Completed 11:58 AM UTC

Commit: bfacd97 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $3.62

fullsend-ai-review[bot]

This comment was marked as outdated.

fullsend-ai-review[bot]

This comment was marked as outdated.

The dispatch workflow no longer checks whether an agent is enabled before
running its setup steps. Instead, `fullsend run` performs the skip: for an
agent with `enabled: false` it exits 0 and relays `skipped=true`,
`disabled=true`, and the skip reason as step outputs.

- Return the typed agentDisabledError from resolveHarnessForLock so the
  disabled-agent error contract matches run.go
- Keep the disabled-agent skip signal in the CLI (disabledSkipResult)
  rather than the pre-script protocol package
- Keep a fix-only enabled check in the Route job so a disabled fix agent
  does not hard-fail fork-PR and PR-fetch checks, and gate Resolve PR head
  on it
- Document the fix-agent exception, the roles workaround limits, the
  reserved `disabled` key in the error semantics table, its migration
  note, and the skipped=true consumer state for disabled agents
- Add behaviour-test steps for disabled agents and status-comment
  assertions

BREAKING CHANGE: `fullsend run` for a disabled agent now exits 0 and
relays skip outputs instead of the workflow blocking the run earlier;
`disabled` is now a reserved pre-script output key.

Assisted-By: rits/zai-org/glm-5-3(pi)
Signed-off-by: Hector Martinez <hemartin@redhat.com>
@fullsend-ai-review

fullsend-ai-review Bot commented Nov 3, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 1:55 PM UTC · Completed 2:14 PM UTC

Commit: 7cc9c31 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $3.14

@fullsend-ai-review fullsend-ai-review Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

See the review comment for full details.

.enabled
" .fullsend/config.yaml 2>"$YQ_ERR" | tail -1) || {
FIX_ENABLED=$(yq '.agents[] | select((.name | downcase) == "fix") | .enabled' .fullsend/config.yaml 2>/dev/null | tail -1) || {
echo "::warning::yq failed checking agents[].enabled — see logs"

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[low] logic-error

In the new fix-check step, yq stderr is redirected to /dev/null, yet the failure handler warns "yq failed checking agents[].enabled — see logs". Nothing diagnostic reaches the logs, and the handler exits 0 so the disabled-fix guard is skipped. The base version captured stderr and printed it in this handler. fullsend run still skips a disabled agent later, so impact is limited to lost diagnostics.

Suggested fix: Capture yq stderr (as the base step did) and print it in the failure handler instead of discarding it.

Comment thread docs/cli/run.md

`fullsend run` terminates early and exits 0 before sandbox creation in these cases:

- **Pre-script skip:** When the harness `pre_script` writes `skipped=true` to `$FULLSEND_PRESCRIPT_OUTPUT`, `fullsend run` logs the skip, emits `skipped=true` (and `reason=<reason>` if provided) to `$GITHUB_OUTPUT` under GitHub Actions, and exits 0.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[low] doc-style

The new ## Early termination and skips section's intro line and both bullets (lines 348-351) are single lines of roughly 80-340 characters. docs/contributing/documentation.md ("Line wrapping") says new Markdown prose is written wrapped at about 80 columns with list continuation lines indented.

Suggested fix: Wrap the intro line and the two list items at about 80 columns, indenting continuation lines to align with the item text.

return res, fmt.Errorf("pre-script output line %d: value for %q contains a control character (%#U) "+
"at offset %d — values must be plain single-line text", lineNum, key, rune(value[i]), i)
}
if key == disabledKey {

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[low] backward-incompatible

disabled is now a reserved pre-script output key: any existing pre-script that emits disabled=... now fails the run with a hard error (prescript.go:191-193), a tightening of the v1 protocol without a version bump. The PR documents it in the error-semantics table and a migration note, and the title carries !. Impact is limited to scripts that already used that key.

Suggested fix: Keep the migration note; optionally call out the reserved key in release notes.

Comment thread internal/cli/run.go
fallbackForgeClient := gh.New(composeGitToken)
harnessPath, fetchDeps, err := resolveAgentSource(ctx, absFullsendDir, agentName, fallbackForgeClient, orgCfg, composeOpts, printer)
if err != nil {
var disabledErr *agentDisabledError

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[low] backward-incompatible

fullsend run for an agent with enabled: false used to exit non-zero; it now exits 0 and relays skipped=true, disabled=true, and reason (run.go:669-683). Callers keyed on the exit code now see success, and disabled non-fix agents now start a workflow run before the CLI skips. Documented and marked with !; consumers must key off skipped.

Suggested fix: Ensure downstream steps and consumers check skipped, not only the exit code. The PR already documents this.

@rh-hemartin

Copy link
Copy Markdown
Member Author

/fs-fix address all feedback, rebase and resolve conflicts

@fullsend-ai-coder

fullsend-ai-coder Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

🤖 Finished Fix · ❌ Failure (post-script /home/runner/work/fullsend/fullsend/.fullsend/.fullsend-cache/resources/sha256/3b4860c748ffd59d27bac1cb5f7f8ca71c8d3b3daba0bbfb012969f8c5127f50/scripts/post-fix.sh failed: exit status 1) · Started 7:18 AM UTC · Completed 7:34 AM UTC

Commit: 7cc9c31 · View workflow run →

Runtime: claude · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $0.58

@fullsend-ai-coder

Copy link
Copy Markdown
Contributor

⚠️ Post-fix script failed — Push rejected (exit code 1)

The fix agent completed, but the post-fix script failed before finishing.

Workflow run: https://github.com/fullsend-ai/fullsend/actions/runs/43173972456

Details:
Could not rebase local 'remove-enabled-gate-from-workflow' onto origin/remove-enabled-gate-from-workflow: the remote branch has commits that conflict with the agent's changes. Resolve the conflict on the PR/MR and re-run /fs-fix.
Rebasing (1/489)
Rebasing (2/489)
Rebasing (3/489)
Rebasing (4/489)
Rebasing (5/489)
Rebasing (6/489)
Rebasing (7/489)
Rebasing (8/489)
Rebasing (9/489)
Rebasing (10/489)
Rebasing (11/489)
Rebasing (12/489)
Rebasing (13/489)
Rebasing (14/489)
Rebasing (15/489)
Rebasing (16/489)
Rebasing (17/489)
Rebasing (18/489)
Rebasing (19/489)
Rebasing (20/489)
Rebasing (21/489)
Rebasing (22/489)
Rebasing (23/489)
Rebasing (24/489)
Auto-merging docs/architecture.md
Auto-merging docs/contributing/config-source-inventory.md
Auto-merging docs/guides/user/jira-integration.md
Auto-merging docs/normative/authorization/v1/README.md
Auto-merging docs/normative/normalized-event/v1/jira-poll-adapter.md
Auto-merging docs/reference/config-reference.md
Auto-merging internal/cli/poll.go
Auto-merging internal/config/defaults.go
Auto-merging internal/config/interfaces.go
Auto-merging internal/jirapoll/poller.go
Auto-merging internal/jirapoll/types.go
CONFLICT (content): Merge conflict in internal/jirapoll/types.go
error: could not apply 177d321... feat(#8401): authorize Jira poll actors without project admin
hint: Resolve all conflicts manually, mark them as resolved with
hint: "git add/rm <conflicted_files>", then run "git rebase --continue".
hint: You can instead skip this commit: run "git rebase --skip".
hint: To abort and get back to the state before "git rebase", run "git rebase --abort".
hint: Disable this message with "git config set advice.mergeConflict false"
Could not apply 177d321... # feat(#8401): authorize Jira poll actors without project admin
Please check the workflow logs for full details and retry with /fs-fix if appropriate.

@jflowers

Copy link
Copy Markdown
Contributor

@rh-hemartin Thanks for reworking this; moving the enabled decision into the CLI looks like the right direction to me. The route job’s yq check reads only .fullsend/config.yaml, while IsAgentExplicitlyDisabled sees the layered config, so the two can drift today.

My earlier findings look addressed, or they no longer apply on main. The scaffold dispatch.yml, repo_overlay_test.go, and the ADR references are all gone there. On your Oct 5 question, I’d still read this as breaking, so the ! and trailer look right to me.

From what I can tell, the CLI skip itself works. The rebase and the cost of starting the stage job may be worth settling before it lands.

Rebase onto main

Wondering if a rebase could come first, since the branch is about 700 commits behind and has conflicts. Since this branched, #8726 added the CEL route step, and its disabled-agent branch in dispatch_route.go returns routeBash. Its comment says the route job’s own agent-enabled checks report these cases. Does that still hold once agent-check is gone?

Gating in the route job through the CLI

Could the route job ask the CLI whether the agent is enabled, for every stage, instead of skipping inside the stage job? The route job can already install the CLI, cached per workflow_sha, so a few things might fall out of that:

one source of truth: layered config, last entry wins, and the same matching as run
no runner, App token, or Podman/OpenShell install for disabled agents
no special case for fix, and no need for the TODO

With retro disabled here, would every PR close in this repo start a runner and mint a token with write scope for nothing? The catch is that the CLI install is currently gated on cel-config and marked continue-on-error, so it would have to run every time. If landing the stage job skip first is easier, I’m fine with that, as long as a separate issue tracks the cost.

fix-check

Is the yq read in fix-check intentional, given it still only sees config.yaml and misses config.base.yaml? It also seems to give enabled: false two different behaviours. fix would skip in the route job with no skipped or disabled outputs, while every other stage skips inside its stage job. Is that split something you want to keep?

The roles paragraph

The roles paragraph in customizing-agents.md says roles no longer gates dispatch, then describes the fullsend and coder role rules. I couldn’t find a role check in reusable-dispatch.yml on main or on this branch, so is the second half still accurate? Would it read cleaner as: roles decides which Apps get installed, and enabled: false turns an agent off?

Version mismatch

What happens when a caller pins fullsend_version to an older CLI but picks up this workflow? From what I can tell, the old CLI fails the job, no disabled output appears, and cleanup posts “Agent run interrupted”. Worth adding a line to the breaking change note saying the workflow and CLI upgrade together?

Coverage for the fix path

The new scenario covers a disabled triage. Is there a reason the disabled fix on a fork PR path has no scenario?

This branch was successfully deployed

1 active (outdated) and 1 inactive deployments
dev — 7cc9c316 Deployed Nov 3, 2026 by rh-hemartin via behaviour #18745
site-preview — bfacd976 Deployed Oct 30, 2026 by github-actions[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

component/dispatch Workflow dispatch and triggers component/runner Agent runner behavior and lifecycle requires-manual-review Review requires human judgment risk/elevated PR risk: elevated tech-debt type/chore Maintenance and housekeeping tasks

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Enabled gate at workflow level can be dropped

3 participants