Skip to content

feat: start runtime workflows from the runner - #8228

Open
waynesun09 wants to merge 16 commits into
feat-workflow-definitions-fetchfrom
feat-workflow-trigger
Open

waynesun09 wants to merge 16 commits into
feat-workflow-definitions-fetchfrom
feat-workflow-trigger

Conversation

@waynesun09

@waynesun09 waynesun09 commented Oct 9, 2026 •

Copy link
Copy Markdown
Member

Summary

Completes ADR 0130 (#8225) on top of #8226: when a harness pins a Claude Code plugin in runtime_workflow:, the runner itself starts the named workflow, /<namespace>:<name> <args>, instead of asking the model to. A pi-extension runtime workflow starts itself from its session hook, so the runner only delivers it. Adds fullsend agent new --runtime-workflow-source/--runtime-workflow-name, a behaviour test with a sample runtime workflow, and two user guides.

Stacked on #8226; review that first.

Naming: the field is runtime_workflow: (a workflow the agent runtime executes itself, not a GitHub Actions workflow). This PR's flags follow it: --runtime-workflow-source and --runtime-workflow-name (spec keys runtime_workflow_source and runtime_workflow_name), the metrics.json key runtime_workflow and the span attribute fullsend.runtime_workflow. The rename is its own commit on top.

Changes

  • Start the workflow (internal/cli/workflow.go, run.go):
    • Claude: every iteration's prompt is the workflow command, with --agent still on the command line. A validation-loop retry runs the same command; with feedback_mode: append the feedback is not passed to the workflow, and the run says so once.
    • pi: the default prompt; the runner starts nothing and skips the Claude-only checks.
    • args expands ${VAR} from the runner environment, so a harness passes the work item (issue ${ISSUE_NUMBER}). Args reach the prompt, the plan and metrics.json, so a credential-shaped variable name, an unset variable, a value that adds a line and a credential-looking value (literal or expanded) are refused; errors name the variable or rule, never the value.
    • Plan-time checks on Claude: the agent's tools: must name Workflow, disallowedTools: must not, its frontmatter must parse, and an enabled tool-allowlist hook whose env.sandbox list omits Workflow is refused.
    • Plan line: Runtime workflow: <source> (sha256:<hash12>) → /<namespace>:<name>[ <args>] on Claude, ... delivered as pi extension on pi.
    • metrics.json gets a runtime_workflow object (source, pin_sha256, kind, and command on Claude) on every runtime, so a dummy run records the launch; the agent span gets fullsend.runtime_workflow.
  • fullsend agent new --runtime-workflow-source <url|path> --runtime-workflow-name <name>: writes the harness runtime_workflow: block and adds Workflow to the agent's tools:; a URL without #sha256= gets a placeholder hash and the command prints how fullsend lock reveals the real one.
  • Allowlist hook test: Workflow events are allowed or denied by FULLSEND_TOOL_ALLOWLIST like any tool.
  • Behaviour test (features/dispatch/runtime-workflow.feature, dummy runtime): commits a sample runtime workflow into the leased repository, runs a harness with runtime_workflow: {source: pipelines/sample-pipeline, name: triage-fanout, args: "issue 1"}, and asserts metrics.json records the command, source, kind and a 64-hex pin. Definition files are removed on cleanup through an optional scm.FileDeleter interface, so scm.Driver is unchanged.
  • Sample runtime workflow (e2e/behaviour/fixtures/workflow/sample-pipeline): a skill that reads an issue's checklist over the GitHub REST API through a helper script, a workflow that checks each item in parallel (issue text passed to agents as delimited data, never as instructions), and a report written to agent-result.json; the sandbox stays read-only and the harness post-script posts the comment.
  • Docs: new docs/guides/user/runtime-workflows.md (run a runtime workflow: local and remote sources, sharing through base:, installing someone else's harness, pi extensions, a troubleshooting table keyed on the exact errors) and docs/guides/user/runtime-workflow-authors.md (make a pipeline repository loadable); updates to harness-reference.md, docs/cli/run.md, docs/cli/agent.md, docs/runtimes.md, docs/runtimes/claude.md, tracing and behaviour-testing guides.

Testing

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

Local run of the sample runtime workflow on the claude runtime (the sandbox image's Claude Code 2.1.258), against public issue #8204, read-only GitHub, post-script in dry-run mode. This run is from before the rename, so its plan line reads Workflow: where this revision prints Runtime workflow::

    Workflow: pipelines/sample-pipeline (sha256:9d2717dc50ed) → /sample-pipeline:triage-fanout issue 8204
...
  ⚙ Workflow
  🧠 The workflow has been launched in the background. It's running with task ID w0x77ctwr. ...
...
  🧠 The workflow has completed! Let me examine the result:
...
  ✓ Agent exited with code 0 (110.3s)
...
post-sample-triage: dry run, not posting
  ✓ Post-script completed (0.4s)
{"source":"pipelines/sample-pipeline","pin_sha256":"9d2717dc50ed46a59bb200b86053b14fa4cc3fb648b45d2672faf5b5aeb56705","kind":"claude-plugin","command":"/sample-pipeline:triage-fanout issue 8204"}

go test -race passes for internal/cli, internal/runtime, internal/harness, internal/agentnew, internal/resolve, internal/security, internal/fetch, internal/lock and pkg/behaviourtest/...; go vet -tags behaviour and -tags playback are clean; the allowlist hook tests pass. The behaviour scenario passed on an earlier revision (35 of 35 scenarios). This revision renames the harness field and the metrics.json key the scenario asserts on, so it needs a re-run.

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

Start harness workflows directly from the runner

✨ Enhancement 🧪 Tests 📝 Documentation 🕐 40+ Minutes

Grey Divider

AI Description

• Start pinned Claude workflows on every iteration instead of relying on the agent to initiate them.
• Validate workflow arguments and agent tool access before sandbox creation; record launches in
 metrics and traces.
• Add workflow scaffolding, a dummy-runtime behaviour scenario, sample definition, and user guides.
Diagram

graph TD
  Config["Pinned definitions"] --> Resolver["Workflow resolver"] --> ToolCheck{"Tool access?"} --> Prompt["Workflow prompt"] --> Claude["Claude runtime"]
  Harness["Harness workflow"] --> Args["Guarded arguments"] --> Resolver
  Prompt --> Metrics["Metrics and traces"]
Loading
High-Level Assessment

Keep the runner-owned prompt approach: it uses the existing workflow resolution and runtime Prompt contract while preserving --agent instructions and tool controls. Having the agent decide whether to start the workflow would not guarantee a launch; adding a separate workflow execution path would duplicate runtime orchestration.

Files changed (60) +2753 / -50

Enhancement (10) +352 / -17
options.goValidate workflow scaffolding options +27/-0

Validate workflow scaffolding options

• Parses definition/name values and rejects malformed workflow specs or incompatible runtimes.

internal/agentnew/options.go

render.goRender workflow-aware agent scaffolds +11/-1

Render workflow-aware agent scaffolds

• Writes the workflow block into generated harnesses and adds Workflow to generated agent tools.

internal/agentnew/render.go

spec.goAccept workflow in agent spec files +2/-0

Accept workflow in agent spec files

• Adds a workflow field to agent-new YAML specifications.

internal/agentnew/spec.go

agent_new.goExpose agent new --workflow and pin reminder +42/-1

Expose agent new --workflow and pin reminder

• Accepts the flag or spec value, reports whether config.yaml pins the definition, and adjusts generated next steps.

internal/cli/agent_new.go

run.goLaunch and record workflows during runs +31/-4

Launch and record workflows during runs

• Expands guarded arguments, checks agent tool access before sandbox creation, supplies the workflow command each iteration, and records metrics and span identity.

internal/cli/run.go

workflow.goBuild guarded workflow commands and launch records +117/-5

Build guarded workflow commands and launch records

• Adds plan-time agent tool checks, retry prompt selection, credential-safe argument expansion, metrics payloads, and trace attributes.

internal/cli/workflow.go

interfaces.goAllow behaviour tests to set workflow entries +5/-0

Allow behaviour tests to set workflow entries

• Adds SetWorkflows to the per-repository config writer so scenarios can change and restore definition entries.

internal/config/interfaces.go

workflow.goReject credential-shaped workflow argument references +67/-6

Reject credential-shaped workflow argument references

• Shares workflow spec validation and refuses credential-shaped environment variables in args when the harness loads.

internal/harness/workflow.go

model.goExpose agent frontmatter tool access +33/-0

Expose agent frontmatter tool access

• Adds an error-reporting reader for allowed and disallowed tools so workflow planning can fail closed.

internal/runtime/model.go

pi_agent.goParse disallowed tools without changing other runtimes +17/-0

Parse disallowed tools without changing other runtimes

• Captures disallowedTools in shared agent parsing while deferring malformed-value errors to the workflow tool-access check.

internal/runtime/pi_agent.go

Tests (23) +1164 / -6
workflow-definition.featureExercise workflow launch planning through dispatch +63/-0

Exercise workflow launch planning through dispatch

• Uses a committed local definition and dummy runtime to assert successful dispatch and recorded workflow command, source, and pin.

e2e/behaviour/features/dispatch/workflow-definition.feature

plugin.jsonIdentify the sample workflow plugin +1/-0

Identify the sample workflow plugin

• Adds a Claude plugin manifest for the sample definition.

e2e/behaviour/fixtures/workflow/sample-pipeline/.claude-plugin/plugin.json

checklist.shAdd sample issue-checklist helper +6/-0

Add sample issue-checklist helper

• Reads an issue through the GitHub API and emits its title and unchecked checklist items as JSON.

e2e/behaviour/fixtures/workflow/sample-pipeline/scripts/checklist.sh

SKILL.mdAdd sample issue-reading skill +18/-0

Add sample issue-reading skill

• Instructs a workflow agent to invoke the plugin's checklist helper and return its JSON result.

e2e/behaviour/fixtures/workflow/sample-pipeline/skills/read-issue/SKILL.md

triage-fanout.jsAdd sample parallel triage workflow +68/-0

Add sample parallel triage workflow

• Defines read, parallel checklist-check, and report phases that produce an agent result.

e2e/behaviour/fixtures/workflow/sample-pipeline/workflows/triage-fanout.js

options_test.goTest workflow option validation +38/-0

Test workflow option validation

• Covers valid and invalid workflow names and Claude-only runtime constraints.

internal/agentnew/options_test.go

render_test.goTest generated workflow harness and tools +35/-0

Test generated workflow harness and tools

• Checks that workflow scaffolding creates a loadable harness and includes Workflow only when requested.

internal/agentnew/render_test.go

spec_test.goTest workflow spec parsing +4/-0

Test workflow spec parsing

• Verifies that agent-new spec files preserve the workflow selection.

internal/agentnew/spec_test.go

agent_new_test.goCover workflow scaffolding CLI paths +97/-0

Cover workflow scaffolding CLI paths

• Tests generation, dry-run and spec input, invalid runtime and syntax, and existing-definition reminders.

internal/cli/agent_new_test.go

workflow_test.goTest workflow launch planning and safeguards +247/-6

Test workflow launch planning and safeguards

• Covers plan output, agent tool refusals, retry prompts, metrics, argument expansion and credential rejection, and trace attributes.

internal/cli/workflow_test.go

workflows_test.goTest workflow config writing and removal +21/-0

Test workflow config writing and removal

• Verifies workflow entries survive marshaling and that setting nil removes the key.

internal/config/workflows_test.go

workflow_test.goTest workflow argument name validation +42/-0

Test workflow argument name validation

• Exercises permitted work-item variables, prohibited credential families, and additional runner-only denials.

internal/harness/workflow_test.go

claude_test.goVerify Claude receives workflow command as prompt +14/-0

Verify Claude receives workflow command as prompt

• Checks that the runtime retains --agent and plugin delivery while replacing its default prompt with the workflow command.

internal/runtime/claude_test.go

model_test.goTest agent tool access parsing +77/-0

Test agent tool access parsing

• Covers string and list forms, inherited tools, malformed entries, and isolation of disallowedTools errors.

internal/runtime/model_test.go

tool_allowlist_pretool_test.pyTest Workflow tool allowlist decisions +20/-0

Test Workflow tool allowlist decisions

• Verifies that the existing pre-tool hook permits or blocks Workflow events according to FULLSEND_TOOL_ALLOWLIST.

internal/security/hooks/tool_allowlist_pretool_test.py

github_test.goTest GitHub SCM file deletion +19/-0

Test GitHub SCM file deletion

• Checks deletion forwarding and not-found handling on a repeated delete.

pkg/behaviourtest/drivers/scm/github/github_test.go

gitlab_test.goTest GitLab SCM file deletion +19/-0

Test GitLab SCM file deletion

• Checks deletion forwarding and not-found handling on a repeated delete.

pkg/behaviourtest/drivers/scm/gitlab/gitlab_test.go

cleanup_test.goExtend cleanup fakes with file deletion +11/-0

Extend cleanup fakes with file deletion

• Adds deletion tracking and stubs needed by workflow fixture cleanup.

pkg/behaviourtest/steps/cleanup_test.go

dispatch_test.goAdapt dispatch SCM fake to deletion contract +4/-0

Adapt dispatch SCM fake to deletion contract

• Adds the DeleteFile stub required by the expanded behaviour SCM interface.

pkg/behaviourtest/steps/dispatch_test.go

fork_test.goAdapt fork SCM fake to deletion contract +4/-0

Adapt fork SCM fake to deletion contract

• Adds the DeleteFile stub required by the expanded behaviour SCM interface.

pkg/behaviourtest/steps/fork_test.go

url_dispatch_test.goAdapt URL-dispatch SCM fake to deletion contract +4/-0

Adapt URL-dispatch SCM fake to deletion contract

• Adds the DeleteFile stub required by the expanded behaviour SCM interface.

pkg/behaviourtest/steps/url_dispatch_test.go

workflow_test.goTest workflow scenario setup, resolution, and cleanup +344/-0

Test workflow scenario setup, resolution, and cleanup

• Covers fixture path safety, partial commits, config restoration, local resolution, and workflow metrics assertions.

pkg/behaviourtest/steps/workflow_test.go

init_test.goTest workflow state reset and SCM fake compatibility +8/-0

Test workflow state reset and SCM fake compatibility

• Checks reset of workflow override fields and adds a file-deletion stub to the suite fake.

pkg/behaviourtest/suite/init_test.go

Documentation (17) +844 / -19
config.tsList workflow guides in site navigation +8/-0

List workflow guides in site navigation

• Adds consumer and author workflow-definition guides to the documentation sidebar.

docs/.vitepress/config.ts

architecture.mdMark runner-owned workflow launch as implemented +1/-1

Mark runner-owned workflow launch as implemented

• Updates the architecture overview to reflect that the runner now starts delivered workflow definitions.

docs/architecture.md

agent.mdDocument agent new workflow scaffolding +40/-0

Document agent new workflow scaffolding

• Explains --workflow, the generated harness and tool list, and the required config.yaml definition pin.

docs/cli/agent.md

run.mdDocument workflow plan and metrics output +13/-7

Document workflow plan and metrics output

• Shows the workflow command in the run plan and describes the workflow object in metrics.json.

docs/cli/run.md

README.mdLink workflow consumer and author guides +2/-0

Link workflow consumer and author guides

• Adds both workflow-definition walkthroughs to the guide index.

docs/guides/README.md

behaviour-drivers.mdDocument SCM file deletion support +1/-1

Document SCM file deletion support

• Notes that behaviour-test SCM drivers can delete files as well as commit them.

docs/guides/dev/behaviour-drivers.md

behaviour-testing.mdExplain workflow behaviour-test coverage +12/-0

Explain workflow behaviour-test coverage

• Documents the new fixture, scenario steps, cleanup, and the dummy runtime's plan-time coverage limits.

docs/guides/dev/behaviour-testing.md

cli-internals.mdList the agent new workflow flag +1/-0

List the agent new workflow flag

• Adds --workflow to the CLI command outline.

docs/guides/dev/cli-internals.md

distributed-tracing.mdDocument workflow span attribute +1/-0

Document workflow span attribute

• Describes the argument-free fullsend.workflow attribute on agent spans.

docs/guides/infrastructure/distributed-tracing.md

bring-your-own-agent.mdDescribe workflow-generated agent files +3/-1

Describe workflow-generated agent files

• Clarifies when scaffolding writes a workflow block and adds Workflow to the agent's tools.

docs/guides/user/bring-your-own-agent.md

workflow-definition-authors.mdAdd workflow-definition authoring guide +338/-0

Add workflow-definition authoring guide

• Covers plugin layout, script metadata, helper skills, pinning, unattended execution, and publishing a reusable harness.

docs/guides/user/workflow-definition-authors.md

workflow-definitions.mdAdd workflow-definition consumer guide +399/-0

Add workflow-definition consumer guide

• Walks through local and remote sources, scaffolding, arguments, locking, running, metrics, CI, and troubleshooting.

docs/guides/user/workflow-definitions.md

config-reference.mdLink workflow configuration walkthrough +2/-1

Link workflow configuration walkthrough

• Connects the workflows configuration reference to the new consumer guide.

docs/reference/config-reference.md

harness-reference.mdSpecify workflow launch and argument rules +11/-7

Specify workflow launch and argument rules

• Documents environment expansion, credential refusals, agent tool checks, retry behaviour, and the command shown in plans.

docs/reference/harness-reference.md

runtimes.mdClarify workflow retry feedback behaviour +1/-1

Clarify workflow retry feedback behaviour

• Notes that workflow retries rerun the command without appended validation feedback.

docs/runtimes.md

claude.mdDescribe Claude workflow execution +9/-0

Describe Claude workflow execution

• Explains plugin delivery, command prompts, retained --agent behaviour, and rejection by other production runtimes.

docs/runtimes/claude.md

runtime.goClarify workflow use of runtime prompts +2/-0

Clarify workflow use of runtime prompts

• Documents that workflow harnesses set RunParams.Prompt to the command on every iteration.

internal/runtime/runtime.go

Other (10) +393 / -8
driver.goAdd file deletion to behaviour SCM contract +3/-0

Add file deletion to behaviour SCM contract

• Introduces DeleteFile for removing scenario-owned definition files from leased repositories.

pkg/behaviourtest/drivers/scm/driver.go

github.goSupport GitHub behaviour fixture deletion +4/-0

Support GitHub behaviour fixture deletion

• Delegates file deletion to the GitHub forge client.

pkg/behaviourtest/drivers/scm/github/github.go

gitlab.goSupport GitLab behaviour fixture deletion +4/-0

Support GitLab behaviour fixture deletion

• Delegates file deletion to the GitLab forge client.

pkg/behaviourtest/drivers/scm/gitlab/gitlab.go

cleanup.goRestore workflow state after behaviour scenarios +15/-0

Restore workflow state after behaviour scenarios

• Restores prior workflows configuration and removes definition files committed by a scenario.

pkg/behaviourtest/steps/cleanup.go

playback.goReuse fixture-tree loading for workflow definitions +34/-5

Reuse fixture-tree loading for workflow definitions

• Separates fixture discovery from commits so workflow scenarios can check paths and record cleanup before writing files.

pkg/behaviourtest/steps/playback.go

registry.goRegister workflow scenario steps +1/-0

Register workflow scenario steps

• Makes the new workflow Gherkin steps available to behaviour scenarios.

pkg/behaviourtest/steps/registry.go

runtime.goAdd agent and workflow metrics scenario support +22/-3

Add agent and workflow metrics scenario support

• Adds runtime-neutral agent-file wording and decodes the workflow object from run metrics.

pkg/behaviourtest/steps/runtime.go

workflow.goImplement workflow-definition scenario steps +294/-0

Implement workflow-definition scenario steps

• Commits checked fixture trees, configures and restores workflow entries, deletes scenario files, and asserts recorded launches.

pkg/behaviourtest/steps/workflow.go

init.goReset workflow configuration scenario state +2/-0

Reset workflow configuration scenario state

• Clears saved workflow override state between behaviour scenarios.

pkg/behaviourtest/suite/init.go

world.goTrack scenario-owned workflow changes +14/-0

Track scenario-owned workflow changes

• Stores the previous workflow entries and committed definition paths for cleanup.

pkg/behaviourtest/world/world.go

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

🤖 Review · ⚠️ Cancelled · Started 3:12 AM UTC · Ended 3:24 AM UTC

Commit: 38a221b · View workflow run →

@qodo-code-review

qodo-code-review Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

Code Review by Qodo

🐞 Bugs (1) 📘 Rule violations (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

Grey Divider


Remediation recommended

1. Failed expansion exposes secret args ✓ Resolved

Resolving commit: ffb48c1

Resolution reason
The finding's flagged code (formatting the original args into a multiline error) no longer exists in
the PR head. expandWorkflowArgs now reports only the variable name for a multiline value: the error
reads "runtime_workflow.args references ${%s}, whose value holds NUL, carriage return or newline
characters..." and the code comment states neither the args nor the value is printed. Literal args
are also scanned up front by CheckWorkflowArgsVariables/checkWorkflowArgsLiteral, and the
post-expansion credential scan error names only the rule, not the text. The test
TestExpandWorkflowArgs_MultiLineValueRefused asserts the args text is not in the error.
Description
expandWorkflowArgs formats the original args into its multiline-value error before running the
secret scan. When args contain both a literal secret and a variable whose value adds a newline, the
returned error includes that secret.

2. Workflow credentials leak into run records ✓ Resolved

Resolving commit: dbdc9af

Description
runAgent calls expandWorkflowArgs only when workflow.args contains ${, so literal arguments
bypass its credential scan. A literal credential in args then reaches the workflow command, which
is shown in the run plan and written to metrics.json.

3. Two guides leave harness undefined ✗ Dismissed
Description
Both new user guides use harness without an inline definition or glossary link at its first
textual occurrence. Readers encounter the term while learning the runtime prerequisites and plugin
naming rules, before the later links to harness field references.

View medium (4)
4. Workflow starts without its work item when a variable is unset ✓ Resolved

Resolving commit: ffb48c1

Resolution reason
Added lookup-based validation to expandWorkflowArgs: variables the lookup cannot resolve are
collected and the run fails with an error naming them, instead of silently expanding to an empty
string and starting the workflow without its work item.
Description
expandWorkflowArgs expands workflow.args with expander, which returns "" for a variable that
is unset or denied, and nothing checks for a missing variable first, unlike runner_env, which goes
through ValidateRunnerEnvWith(lookup). When ISSUE_NUMBER (or any other referenced variable) is
missing from the runner environment, issue ${ISSUE_NUMBER} becomes issue  and the runner starts
/<ns>:<name> issue  without the work item, then records that command in the plan and metrics.json
as if the run succeeded normally.

5. Restricted workflows fail after planning ✓ Resolved

Resolving commit: ffb48c1

Resolution reason
Added checkWorkflowToolAllowlist, invoked in runAgent alongside checkWorkflowAgentTools, which fails
the run at plan time when the tool allowlist hook is enabled and FULLSEND_TOOL_ALLOWLIST in
env.sandbox does not name Workflow.
Description
checkWorkflowAgentTools checks the agent frontmatter but not an enabled tool allowlist. If that
allowlist omits Workflow, the new plan-time check passes and the security hook blocks the workflow
tool invocation during execution.

6. Consumer setup lacks numbered steps ✓ Resolved

Resolving commit: dbdc9af

Description
The author guide gives commands for obtaining the consumer's file and tree hashes as unordered
bullets rather than an ordered procedure. Readers following the installation example must infer the
sequence from the prose and bullet labels.

7. Sample issue reads bypass forge clients ✗ Dismissed
Description
checklist.sh reads an issue with a direct gh api invocation instead of a forge.Client
operation. When the sample workflow runs its read-issue skill, its GitHub request bypasses the forge
abstraction used by the behaviour-test SCM drivers.


Informational

8. Credential-name doc comment lists a rule the code skips ✓ Resolved

Resolving commit: ffb48c1

Resolution reason
The doc comment was rewritten to describe the KEY rule as "a KEY segment after a
credentialKeyQualifiers segment" and to state that "A bare *_KEY such as ISSUE_KEY is allowed",
removing the claim that *_KEY is matched.
Description
The doc comment on CredentialShapedEnvName says it matches *_KEY. credentialEnvSuffixes leaves
out bare _KEY on purpose so ISSUE_KEY passes, as the comment just above it explains. A later
maintainer who trusts the function doc (the commit message repeats *_KEY) could "fix" the code to
match it, which would start refusing Jira ${ISSUE_KEY} args, or could assume names like FOO_KEY
are already blocked.

9. Failed cleanup affects later scenarios 🐞 Bug ☼ Reliability
Description
CleanupScenario logs an exhausted workflow-config restoration error and continues, while
RemoveWorkflowDefinitionFiles likewise logs deletion failures and discards its list of paths. When
either repository operation keeps failing, the next scenario can encounter a stale workflow entry or
an existing definition path, and ValidateSlotClean checks for neither.
Code

pkg/behaviourtest/steps/cleanup.go[R301-305]

+	if w.WorkflowsOverridden {
+		if err := cleanupRetry(w.Logf, "restore workflows", func() error {
+			return RestoreWorkflows(w)
+		}); err != nil {
+			worldLogf(w, "behaviour cleanup: restore workflows: %v", err)
Relevance

● Weak

A recent, closely matching cleanup precedent explicitly rejected scenario-level changes after retry
exhaustion.

PR-#6217

ⓘ Recommendations generated based on similar findings in past PRs

Evidence
Both new cleanup operations continue after their retry budget is exhausted, while the slot
cleanliness check examines only unrelated configuration fields.

pkg/behaviourtest/steps/cleanup.go[298-311]
pkg/behaviourtest/steps/workflow.go[120-135]
pkg/behaviourtest/steps/workflow.go[242-257]
pkg/behaviourtest/steps/cleanup.go[42-68]

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

## Issue description
Failed workflow cleanup is logged but does not prevent reuse of the affected leased repository.
## Fix Focus Areas
- pkg/behaviourtest/steps/cleanup.go[298-311]
- pkg/behaviourtest/steps/workflow.go[120-135]
- pkg/behaviourtest/steps/cleanup.go[42-68]
## Recommended Fix
Propagate workflow restoration and deletion failures to slot management so an unclean repository cannot be leased again; retain failed paths for any subsequent retry.

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

Dismiss ↗ | View ↗


Grey Divider

Context sources
✅ Compliance rules (platform): 89 rules
✅ Skills: 5 invoked
  writing-user-docs
  writing-adrs
  check-patch-coverage
  cutting-releases
  adr-corner
✅ Cross-repo context — repo relationships
  Explored: repo: fullsend-ai/.fullsend (sha: 7c163cad) — View relationship
  Explored: repo: fullsend-ai/agents (sha: f3dd8897) — View relationship
  Explored: repo: fullsend-ai/mlflow-assessments (sha: 2fdb0120) — View relationship
Review mode: Auto: 🧠 Deep: Dense cross-cutting workflow, security, runtime, CLI, and behavior changes.

Grey Divider

Tip of the day
💡 Did you know, you can import rules and skills only from the repositories your team follows

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

Grey Divider

Qodo Logo

Comment thread internal/cli/run.go Resolved
Comment thread internal/cli/workflow.go Resolved
Comment thread internal/cli/run.go Resolved
Comment thread internal/cli/run.go Resolved
Comment thread internal/harness/workflow.go Resolved
@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

Site preview

Preview: https://6c27abd3-site.fullsend-ai.workers.dev

Commit: 064ca080f0a32b262055266348d5e8ca173c2e61

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 3:26 AM UTC · Completed 3:42 AM UTC

Commit: 7d9635c · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $5.48

@waynesun09

Copy link
Copy Markdown
Member Author

/fs-fix-stop

@github-actions github-actions Bot added the fullsend-no-fix Skip bot-triggered fix agent runs label Oct 9, 2026
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown

Fix agent disabled for this PR. Remove the fullsend-no-fix label or use /fs-fix to re-engage.

@fullsend-ai-review fullsend-ai-review Bot added the risk/moderate PR risk: moderate label Oct 9, 2026
@fullsend-ai-review

fullsend-ai-review Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

Risk Assessment: moderate (2/5)

Details

Large additive feature with high file count and churn in core runner files, mitigated by strictly opt-in workflow configuration, zero protected or security-sensitive changes, clean dependency footprint, and close alignment with accepted ADR 0130.

Previous run

Risk Assessment: moderate (2/5)

Details

Large additive feature with high file count and churn in core runner files, mitigated by strictly opt-in workflow configuration, zero protected or security-sensitive changes, clean dependency footprint, and close alignment with accepted ADR 0130.

Previous run (2)

Risk Assessment: moderate (2/5)

Details

Large additive feature with high file count and churn in core runner files, mitigated by strictly opt-in workflow configuration, zero protected or security-sensitive changes, clean dependency footprint, and close alignment with accepted ADR 0130.

Previous run (3)

Risk Assessment: moderate (2/5)

Details

Large additive feature with high file count and churn in core runner files, mitigated by strictly opt-in workflow configuration, zero protected or security-sensitive changes, clean dependency footprint, and close alignment with accepted ADR 0130.

Previous run (4)

Risk Assessment: moderate (2/5)

Details

Large additive feature with high file count and churn in core runner files, mitigated by strictly opt-in workflow configuration, zero protected or security-sensitive changes, clean dependency footprint, and close alignment with accepted ADR 0130.

Previous run (5)

Risk Assessment: moderate (2/5)

Details

Large additive feature with high file count and churn in core runner files, mitigated by strictly opt-in workflow configuration, zero protected or security-sensitive changes, clean dependency footprint, and close alignment with accepted ADR 0130.

@fullsend-ai-review

fullsend-ai-review Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

Review

Findings

Low

  • [logic-error] internal/cli/workflow.go:362 — Expansion is enabled by the presence of ${, but os.Expand also interprets bare $VAR and $1, so adding ${ISSUE_NUMBER} changes previously literal arguments (issue ${ISSUE_NUMBER} budget $5 fails with an unset-variable error for 5).
    Remediation: Expand only the documented ${VAR} syntax and add a mixed interpolation/literal-dollar regression test.
  • [api-contract] docs/guides/user/runtime-workflows.md:389 — The success check looks for a Workflow: line, but the runner prints Runtime workflow: (internal/cli/run.go:1507).
    Remediation: Use Runtime workflow: consistently in the success check and its absence diagnostic.
  • [incomplete-doc] docs/runtimes/pi.md:183 — The pi runtime page does not document runtime_workflow support (pi extension delivered, started from the session hook).
    Remediation: Add a Runtime workflows row to the "At a glance" table and describe how pi loads runtime_workflow resources.

Earlier findings

  • docs/guides/user/runtime-workflows.md:400, api-contract, reclassified to info — wording is consistent with the run directory layout in run.go.
  • internal/harness/workflow.go:102, backward-incompatible, reclassified to info — the cited checks are unchanged context; only the args-variable check is new, on an unreleased surface.
  • internal/agentnew/options.go:62, error-handling-idiom, reclassified to info — the file mixes %s and %q; cosmetic.
  • internal/runtime/pi_agent.go:104, error-handling-idiom, reclassified to info — the trimmed prefix originates in the same file.
  • docs/reference/config-reference.md:55, incomplete-doc, reclassified to info — the allowlist check predates this PR.
  • docs/guides/infrastructure/layered-config-reference.md:426, incomplete-doc, reclassified to info — the allowlist check predates this PR.

Labels: Feature PR adding runtime-workflow start logic with behaviour-test (e2e) coverage.


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

Previous run

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

Review

Findings

Low

  • [logic-error] internal/cli/workflow.go:362 — os.Expand also expands bare $VAR and single-character references such as $5 once args contain one braced reference, so issue ${ISSUE_NUMBER} budget $5 fails on an unset variable 5. harness.CheckWorkflowArgsVariables behaves the same way.
    Remediation: Expand only braced ${VAR} references and add a test combining a braced reference with literal $5 or $NAME text.
  • [api-contract] docs/guides/user/runtime-workflows.md:389 — The guide's prose names a Workflow: plan line (lines 257, 389, 393), but the runner prints Runtime workflow (internal/cli/run.go:1507) and the guide's own examples show Runtime workflow:.
    Remediation: Change the prose references to the Runtime workflow: line.
  • [error-handling-idiom] internal/agentnew/options.go:62 — The --runtime-workflow-name value is formatted with %s instead of %q (also at line 145), unlike the %q quoting Validate() uses.
    Remediation: Use %q in both messages.
  • [error-handling-idiom] internal/runtime/pi_agent.go:104 — DisallowedErr is built by trimming a prefix off err.Error(), which breaks silently if the piToolSpecs wording changes.
    Remediation: Give piToolSpecs a field-name parameter or return a structured error.
  • [backward-incompatible] internal/harness/workflow.go:104 — runtime_workflow.args was literal in feat: fetch, lock and deliver runtime workflows #8226 and now expands ${...} from the runner environment and refuses credential-shaped names. The field is from the unreleased stacked predecessor, so the risk is small and depends on release ordering.
    Remediation: Merge and release feat: fetch, lock and deliver runtime workflows #8226 and this PR together, or note the new expansion in the release notes.
  • [incomplete-doc] docs/reference/config-reference.md:55 — The allowed_remote_resources list omits runtime workflows (runtime_workflow.source). The gap comes from the stacked predecessor feat: fetch, lock and deliver runtime workflows #8226; this PR does not touch the file.
    Remediation: Add runtime workflows to the list, in this stack or as a follow-up.
  • [incomplete-doc] docs/guides/infrastructure/layered-config-reference.md:426 — The allowed_remote_resources section omits runtime workflows; same origin as above.
    Remediation: Add runtime workflows to the parenthetical list.
  • [incomplete-doc] docs/runtimes/pi.md:183 — This PR links the new runtime-workflows guide from docs/runtimes/claude.md but docs/runtimes/pi.md has no mention of runtime_workflow: (no 'At a glance' entry, no 'See also' link).
    Remediation: Note that runtime_workflow: definitions are delivered as pi extensions and link the guide.

Labels: PR adds runner-side start of runtime workflows, a harness field behaviour, CLI flags and user guides


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

Previous run (2)

<!-- fullsend:review-findings-v2:eyJ2ZXJzaW9uIjoyLCJhY3Rpb24iOiJhcHByb3ZlIiwiZmluZGluZ3MiOlt7InNldmVyaXR5IjoibG93IiwiY2F0ZWdvcnkiOiJsb2dpYy1lcnJvciIsImZpbGUiOiJpbnRlcm5hbC9jbGkvd29ya2Zsb3cuZ28iLCJpZCI6ImZfYTU3MTliNjQ3NmZhNzdmZCIsImxpbmUiOjM1OSwiYWN0aW9uYWJsZSI6ZmFsc2V9LHsic2V2ZXJpdHkiOiJsb3ciLCJjYXRlZ29yeSI6ImFwaS1jb250cmFjdCIsImZpbGUiOiJkb2NzL2d1aWRlcy91c2VyL3J1bnRpbWUtd29ya2Zsb3dzLm1kIiwiaWQiOiJmXzM2MDVkYzA3OWM5ZmY3MzgiLCJsaW5lIjo0MDAsImFjdGlvbmFibGUiOmZhbHNlfSx7InNldmVyaXR5IjoibG93IiwiY2F0ZWdvcnkiOiJlcnJvci1oYW5kbGluZy1pZGlvbSIsImZpbGUiOiJpbnRlcm5hbC9hZ2VudG5ldy9vcHRpb25zLmdvIiwiaWQiOiJmXzNlNGI0MGJhMzczOTViODIiLCJsaW5lIjo2MiwiYWN0aW9uYWJsZSI6ZmFsc2V9LHsic2V2ZXJpdHkiOiJpbmZvIiwiY2F0ZWdvcnkiOiJlcnJvci1oYW5kbGluZy1pZGlvbSIsImZpbGUiOiJwa2cvYmVoYXZpb3VydGVzdC9zdGVwcy93b3JrZmxvdy5nbyIsImlkIjoiZl82YzAxNzAyNmMwNDNiYTM0IiwibGluZSI6OTMsImFjdGlvbmFibGUiOmZhbHNlfSx7InNldmVyaXR5IjoiaW5mbyIsImNhdGVnb3J5IjoiYXBpLXNoYXBlIiwiZmlsZSI6InBrZy9iZWhhdmlvdXJ0ZXN0L3N0ZXBzL3dvcmtmbG93LmdvIiwiaWQiOiJmXzU0ODJiZTkwZTRkYzJmNDMiLCJsaW5lIjoxMjQsImFjdGlvbmFibGUiOmZhbHNlfSx7InNldmVyaXR5IjoibG93IiwiY2F0ZWdvcnkiOiJlcnJvci1oYW5kbGluZy1pZGlvbSIsImZpbGUiOiJpbnRlcm5hbC9ydW50aW1lL3BpX2FnZW50LmdvIiwiaWQiOiJmX2YzYjllYTg3MTYxNGM5OGEiLCJsaW5lIjoxMDAsImFjdGlvbmFibGUiOmZhbHNlfSx7InNldmVyaXR5IjoibWVkaXVtIiwiY2F0ZWdvcnkiOiJzZWNyZXQtZXhwb3N1cmUiLCJmaWxlIjoiaW50ZXJuYWwvaGFybmVzcy93b3JrZmxvdy5nbyIsImlkIjoiZl85NWM2NWQ0N2M0MTRlNTY3IiwibGluZSI6MjQ1fSx7InNldmVyaXR5IjoibWVkaXVtIiwiY2F0ZWdvcnkiOiJwcm9tcHQtaW5qZWN0aW9uIiwiZmlsZSI6ImUyZS9iZWhhdmlvdXIvZml4dHVyZXMvd29ya2Zsb3cvc2FtcGxlLXBpcGVsaW5lL3dvcmtmbG93cy90cmlhZ2UtZmFub3V0LmpzIiwiaWQiOiJmXzJiOGZiZWU4OTlhM2Y4NzciLCJsaW5lIjozMn0seyJzZXZlcml0eSI6ImxvdyIsImNhdGVnb3J5IjoibWlzc2luZy1hdXRob3JpemF0aW9uIiwiZmlsZSI6ImRvY3MvYXJjaGl0ZWN0dXJlLm1kIiwiaWQiOiJmX2I0OTVjYjRiNTRhYTM4MjEiLCJsaW5lIjoyNzAsImFjdGlvbmFibGUiOmZhbHNlfV0sImRpc3Bvc2l0aW9ucyI6W3siaWQiOiJmX2E1NzE5YjY0NzZmYTc3ZmQiLCJzdGF0dXMiOiJvcGVuIn0seyJpZCI6ImZfMzYwNWRjMDc5YzlmZjczOCIsInN0YXR1cyI6Im9wZW4ifSx7ImlkIjoiZl9iNDk1Y2I0YjU0YWEzODIxIiwic3RhdHVzIjoicmVzb2x2ZWRfYnlfY2hhbmdlIn0seyJpZCI6ImZfM2U0YjQwYmEzNzM5NWI4MiIsInN0YXR1cyI6Im9wZW4ifSx7ImlkIjoiZl82YzAxNzAyNmMwNDNiYTM0Iiwic3RhdHVzIjoicmVjbGFzc2lmaWVkIn0seyJpZCI6ImZfNTQ4MmJlOTBlNGRjMmY0MyIsInN0YXR1cyI6InJlY2xhc3NpZmllZCJ9LHsiaWQiOiJmXzk1YzY1ZDQ3YzQxNGU1NjciLCJzdGF0dXMiOiJyZXNvbHZlZF9ieV9jaGFuZ2UifSx7ImlkIjoiZl8yYjhmYmVlODk5YTNmODc3Iiwic3RhdHVzIjoicmVzb2x2ZWRfYnlfY2hhbmdlIn1dfQ== -->

Review

Findings

Low

  • [logic-error] internal/cli/workflow.go:359 — Malformed variable references silently change the workflow arguments instead of failing validation. Go's os.Expand consumes invalid ${... syntax without invoking the lookup callback, so issue ${ISSUE_NUMBER becomes issue ISSUE_NUMBER and ${} disappears. Neither CheckWorkflowArgsVariables nor the unset-variable check catches these cases, so a mistyped work-item argument can reach the workflow.
    Remediation: Validate expansion syntax before calling os.Expand and reject unterminated or empty braced references. Add tests covering ${ISSUE_NUMBER and ${}.
  • [api-contract] docs/guides/user/runtime-workflows.md:400 — Earlier finding, still open: the metrics.json runtime_workflow example and field table in this section are an API-contract surface. This push rewrote history, so the delta since the prior review could not be enumerated, and nothing in the current diff shows this finding was addressed. Check the example and table against workflowLaunch in internal/cli/workflow.go. For a source pinned through a URL base:, source is the resolved tree URL, which the table does not mention.
  • [error-handling-idiom] internal/agentnew/options.go:62 — ParseRuntimeWorkflow formats the user-supplied --runtime-workflow-name with %s instead of %q.
    Remediation: Quote the user-supplied value with %q.
  • [error-handling-idiom] internal/runtime/pi_agent.go:100 — parsePiAgent builds def.DisallowedErr by trimming another error's message with strings.TrimPrefix(err.Error(), "agent definition: tools "). The prefix is hard-coded in piToolSpecs' error strings, so rewording them would silently break the trim.
    Remediation: Pass the field name into piToolSpecs (or a helper) so each error is formatted for its own field without parsing error text.

Earlier findings

  • f_b495cb4b54aa3821 — resolved_by_change: docs/architecture.md now says fetch, lock, delivery and start are implemented, which matches ADR 0130 and this PR stacked on feat: fetch, lock and deliver runtime workflows #8226.
  • f_6c017026c043ba34 — reclassified to info: errNoFileDeleter is used in two places (pkg/behaviourtest/steps/workflow.go, returned and logged), so a package-level error variable is ordinary Go.
  • f_5482be90e4dc2f43 — reclassified to info: removeRuntimeWorkflowFiles uses the shared cleanupRetry and worldLogf, and CleanupScenario already has other per-item cleanup loops inline, so this is a style preference.
Previous run (3)

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

Review

Findings

Low

  • [logic-error] internal/cli/workflow.go:359 — expandWorkflowArgs runs os.Expand over the whole args string as soon as it contains ${, so a bare $5 elsewhere in the string is also treated as a variable reference. issue ${ISSUE_NUMBER} budget $5 fails with an unset variable 5, while issue 42 budget $5 passes through unchanged. It fails closed, but it contradicts the doc comment ("expands ${VAR}"). Remediation: expand only braced references and add a regression test that mixes ${ISSUE_NUMBER} with literal dollar text.
  • [api-contract] docs/guides/user/runtime-workflows.md:400 — The documented inspection command is jq .workflow .../metrics.json, but the runner writes the object under runtime_workflow (internal/cli/run.go), so the command prints null. Remediation: change it to jq .runtime_workflow ....
  • [missing-authorization] docs/architecture.md:270 — The PR links no tracking issue. The work traces to ADR 0130 (docs(adr): make runtime workflows harness resources #8225) and feat: fetch, lock and deliver runtime workflows #8226, which are both PRs. Remediation: link a tracking issue in the PR description.
  • [error-handling-idiom] internal/agentnew/options.go:62 — Carried from the prior review. This run did not re-evaluate it because the style-conventions sub-agent timed out. The code at the cited line is still in the PR head.
  • [error-handling-idiom] pkg/behaviourtest/steps/workflow.go:93 — Carried from the prior review. This run did not re-evaluate it because the style-conventions sub-agent timed out. The code at the cited line is still in the PR head.
  • [api-shape] pkg/behaviourtest/steps/workflow.go:124 — Carried from the prior review. This run did not re-evaluate it because the style-conventions sub-agent timed out. The code at the cited line is still in the PR head.
  • [sub-agent-failure] N/A — The style-conventions and docs-currency sub-agents timed out and returned no findings. Style and documentation staleness checks are incomplete for this run, apart from the docs finding above, which the correctness sub-agent reported.

Earlier findings

  • f_95c65d47c414e567 — resolved_by_change: the security sub-agent found that credential-shaped and runner-only variable references are refused. Literal and expanded args are also scanned before launch.
  • f_2b8fbee899a3f877 — resolved_by_change: triage-fanout.js now wraps issue-derived text as JSON inside a labelled <untrusted-issue-content> block and tells the agents to treat it as data. The read-issue skill marks the helper output as untrusted.

Re-review note: No findings at or above the effective blocking threshold remain open. New low/info findings do not block this review or start a fix run.

Previous run (4)

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

Review

Findings

Medium

  • [secret-exposure] internal/harness/workflow.go:245 — CredentialShapedEnvName matches only a fixed list of suffixes (_TOKEN, _PASSWORD, _API_KEY, ...), _SECRET-containing names and OTEL_*. DB_PASSWD, DB_PASS, HTTP_AUTHORIZATION and GH_TOKEN_B64 pass, and harnessEnvLookup refuses only the OIDC and provider-only keys. A bare opaque password or base64 token does not match the value scanner either, so args: issue ${ISSUE_NUMBER} ${DB_PASSWD} would reach the prompt, the run plan and metrics.json. It needs the trusted harness author to write that args string, so this is a guardrail gap, not a trust-boundary break.
    Remediation: Prefer an explicit allowlist of runner variables permitted in workflow.args (e.g. ISSUE_NUMBER, REPO_FULL_NAME). At minimum also refuse names ending in PASS, PASSWD, AUTH, AUTHORIZATION and credential suffixes followed by another suffix (_TOKEN_*), with tests.

  • [prompt-injection] e2e/behaviour/fixtures/workflow/sample-pipeline/workflows/triage-fanout.js:32 — The Read agent receives issue titles and checklist items written by the issue author through checklist.sh, but neither its prompt nor read-issue/SKILL.md marks that output as untrusted. untrusted() is applied only in the later Check and Report phases, which cannot undo manipulation of the first agent. The fixture is also the copyable reference example in workflow-definitions.md and workflow-definition-authors.md.
    Remediation: Add an explicit untrusted-data instruction to the Read prompt and the skill. Better, parse the issue deterministically in the workflow script and pass the content only through untrusted().

Low

  • [logic-error] internal/cli/workflow.go:359 — Expansion is inconsistent for bare dollar expressions. Args without ${ are left alone, but once any ${VAR} appears os.Expand also expands bare $VAR and positional names, in both expandWorkflowArgs and harness.CheckWorkflowArgsVariables (internal/harness/workflow.go:262-266). $FILTER alone is literal, while $FILTER ${ISSUE_NUMBER} looks up FILTER and fails if it is unset. The docs describe only ${VAR}. Fail-closed, with no credential leak.
    Remediation: Use one shared parser that recognises only ${VAR} for validation and expansion, or document the behaviour. Add tests for mixed braced and bare references.

  • [error-handling-idiom] internal/agentnew/options.go:62 — Error messages at lines 62, 144 and 147 format user-provided or enumerated values with %s. docs/contributing/go-code.md requires %q, and sibling errors in the file use it.
    Remediation: Use %q for name, o.Workflow.Name and o.Runtime, and update the assertions in options_test.go and agent_new_test.go.

  • [error-handling-idiom] pkg/behaviourtest/steps/workflow.go:93 — Sentinel errNoFileDeleter is built with fmt.Errorf and has no format verbs; the repo's Go guide builds sentinels with errors.New.
    Remediation: Use errors.New(...).

  • [api-shape] pkg/behaviourtest/steps/workflow.go:124 — RemoveWorkflowDefinitionFiles is exported so CleanupScenario can call it, but both are in package steps (only caller: cleanup.go:302). The export widens the package's public API.
    Remediation: Unexport it and update the caller, tests and doc comment.


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)

Review

Findings

High

  • [breaking-change-unmarked] pkg/behaviourtest/drivers/scm/driver.go:33 — The PR adds a required method, DeleteFile(ctx, owner, repo, path, message string) error, to the exported scm.Driver interface. Any scm.Driver implementation or test fake outside this repo will stop compiling. COMMITS.md § Breaking changes requires a ! suffix on the commit message and PR title, plus a BREAKING CHANGE: trailer. The title "feat: start workflow definitions from the runner" has no !. Earlier additions to this interface (ci: add dummy-playback behavior scenarios and fixtures #7941, ci: extract shared behaviour-test dispatch and SCM infrastructure #7993) did carry !. docs/guides/dev/behaviour-testing.md has a migration note for each earlier scm.Driver addition, such as DeleteRepo, ListPullRequestReviews and GetFileContentAtRef. The PR head copy of that guide never mentions DeleteFile.
    Remediation: Either make DeleteFile an optional interface (for example scm.FileDeleter) and type-assert for it in pkg/behaviourtest/steps/cleanup.go, as ci.RunLister does so external drivers keep compiling. Or add ! to the PR title and commit message, add a BREAKING CHANGE: trailer, and add a scm.Driver.DeleteFile migration note to docs/guides/dev/behaviour-testing.md.

Medium

  • [prompt-injection] e2e/behaviour/fixtures/workflow/sample-pipeline/workflows/triage-fanout.js:43 — The sample puts issue titles and checklist items, written by outside users, straight into the child agents' instructions (Issue "${issue.title}", checklist item ${i + 1}: "${item}"). Nothing marks them as untrusted data. The report agent's prompt also includes item and the model-produced c.reason, via JSON.stringify inside a free-text instruction. JSON.stringify encodes the data but does not stop injection. An issue author can embed instructions that skew the readiness judgments or the report. The new user guides present this fixture as the template authors should copy.
    Remediation: Keep the task instructions separate from clearly delimited, untrusted issue data, and tell the evaluators not to follow instructions found inside that data. Have the script write agent-result.json itself instead of asking an agent to write it. Failing that, check the written object against the report the script computed. Add adversarial title and item cases.

Low

  • [secret-exposure] internal/cli/run.go:1121 — The credential-value check only runs when workflow.args contains ${. harness.CheckWorkflowArgsVariables returns early without ${, and expandWorkflowArgs (which holds the SecretRedactor scan) is not called. A literal credential in args, for example a ghp_ token, therefore passes ValidateWorkflowSpec and reaches the printed plan, the agent prompt and metrics.json. The value can only come from the harness YAML, so this is defence in depth against operator error, not an attacker-controlled path.
    Remediation: Run the SecretRedactor value check on every non-nil workflow's args, even when nothing needs expanding. Add a regression test that a literal credential is refused and never printed.

  • [error-handling idioms] internal/cli/workflow.go:76 — checkWorkflowAgentTools formats the user-supplied agentName and h.Agent with %s on lines 76, 79 and 82. checkWorkflowRuntime in the same file uses %q, and docs/contributing/go-code.md ("Use %q for values in error messages") prescribes it. Changing it means updating the matching assertions in workflow_test.go, the troubleshooting table in docs/guides/user/workflow-definitions.md, and docs/reference/harness-reference.md (line 222).
    Remediation: Use %q for these identifiers and update the dependent tests and docs.


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.

@rh-hemartin

Copy link
Copy Markdown
Member

Be careful with implementing before the ADR is accepted. I'm fighting very hard against "sunken cost fallacy", so keep in mind this can be wasted tokens and budget.

@waynesun09
waynesun09 force-pushed the feat-workflow-definitions-fetch branch 2 times, most recently from 5869bda to 4010cbc Compare October 9, 2026 13:43
@waynesun09
waynesun09 force-pushed the feat-workflow-definitions-fetch branch from 4010cbc to 72370a8 Compare October 9, 2026 14:33
@waynesun09
waynesun09 force-pushed the feat-workflow-trigger branch from 7d9635c to dbdc9af Compare October 9, 2026 16:32
A harness added by URL now resolves a relative runtime_workflow.source in
its own repository at the commit it was fetched from (#8226), so the
guides stop calling it refused, the troubleshooting row names the
branch-ref error, and the quoted base error matches its new wording.

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>
@fullsend-ai-review

fullsend-ai-review Bot commented Nov 11, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ⚠️ Cancelled · Started 7:55 PM UTC · Completed 8:01 PM UTC

Commit: ffb48c1 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $0.92

This review was superseded by a newer commit. Check the current pull request HEAD's review status before merging.

@fullsend-ai-review fullsend-ai-review Bot removed the ready-for-merge All reviewers approved — ready to merge label Nov 11, 2026
Pin the remote example to 7068965, the pushed commit that last changed
e2e/behaviour/fixtures/workflow/sample-pipeline, and replace every pasted
output with a fresh run of this branch's binary: agent new, the three
lock steps and lock.yaml (now field: runtime_workflow), the remote plan
line, the local run against issue 8204, and jq .runtime_workflow. The
tree hash is 99be97667c85 for both the path and the remote source.

Also count the tool allowlist hook's Workflow check: fullsend run makes
five plan-time checks, not four.

Assisted-by: Claude
Signed-off-by: Wayne Sun <gsun@redhat.com>
@fullsend-ai-review

fullsend-ai-review Bot commented Nov 11, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 8:03 PM UTC · Completed 8:26 PM UTC

Commit: 064ca08 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $4.59

fullsend-ai-review[bot]

This comment was marked as outdated.

@fullsend-ai-review
fullsend-ai-review Bot dismissed their stale review November 11, 2026 20:26

Superseded by updated review

@fullsend-ai-review fullsend-ai-review Bot added ready-for-merge All reviewers approved — ready to merge component/runner Agent runner behavior and lifecycle component/harness Agent harness, config, and skills loading component/docs User-facing documentation labels Nov 11, 2026
@fullsend-ai-review

fullsend-ai-review Bot commented Nov 11, 2026 •

Copy link
Copy Markdown

🤖 Finished Review · ✅ Success · Started 9:04 PM UTC · Completed 9:16 PM UTC

Commit: 064ca08 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $2.43

@fullsend-ai-review fullsend-ai-review Bot removed the ready-for-merge All reviewers approved — ready to merge label Nov 11, 2026
Comment thread internal/cli/workflow.go Resolved
Comment thread docs/guides/user/runtime-workflows.md Resolved
@fullsend-ai-review
fullsend-ai-review Bot dismissed their stale review November 11, 2026 21:16

Superseded by updated review

@fullsend-ai-review fullsend-ai-review Bot added ready-for-merge All reviewers approved — ready to merge type/feature New capability request component/e2e End-to-end tests labels Nov 11, 2026

This branch was successfully deployed

1 active and 1 inactive deployments
dev — 064ca080 Deployed Nov 11, 2026 by waynesun09 via behaviour #19737
site-preview — 064ca080 Deployed Nov 11, 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/docs User-facing documentation component/e2e End-to-end tests component/harness Agent harness, config, and skills loading component/runner Agent runner behavior and lifecycle fullsend-no-fix Skip bot-triggered fix agent runs ready-for-merge All reviewers approved — ready to merge risk/moderate PR risk: moderate type/feature New capability request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants