[rh-hemartin]

Moves fullsend/agents/experiments repo grants out of the general triagers/writers teams into dedicated fullsend-group teams (fullsend-triagers, fullsend-writers), matching the /fs-triage and /fs-code access for the fullsend group.

Lint passes (python3 scripts/lint.py).

Assisted-By: glm-5-3(pi)

[qodo-code-review]

PR Summary by Qodo

Separate Fullsend repository access into dedicated teams

⚙️ Configuration changes ✨ Enhancement 🕐 Less than 10 minutes

AI Description

• Add dedicated triage and write teams for fullsend, agents, and experiments.
• Remove fullsend and agents grants from general teams, separating future access management.
• Match the new teams’ initial rosters to the corresponding general teams.

Diagram

graph TD
  Config["peribolos.yaml"] --> Triagers["General triagers"] --> Other["Other repos"]
  Config --> Writers["General writers"] --> Other
  Config --> FSTriagers["Fullsend triagers"] --> FSRepos["Fullsend group repos"]
  Config --> FSWriters["Fullsend writers"] --> FSRepos

Loading

High-Level Assessment

Dedicated teams are the appropriate way to manage Fullsend-group access independently. Keeping grants on the general teams would not provide that separation.

Files changed (1) +44 / -4

Other (1) +44 / -4

peribolos.yamlDefine dedicated Fullsend triage and write teams +44/-4

Define dedicated Fullsend triage and write teams

• Removes fullsend and agents grants from the general triagers and writers teams. Adds closed fullsend-triagers and fullsend-writers teams with matching initial rosters and triage or write grants to fullsend, agents, and experiments.

peribolos.yaml

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

Tip of the day

💡 Did you know, you can type 'qodo, fix this' on a finding and the fix lands right on your PR

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 11:58 AM UTC · Completed 12:10 PM UTC

Commit: c596d71 · View workflow run →

Runtime: claude · Model: opus → claude-opus-5-5 · Effort: high · Cost: $2.40

[fullsend-ai-retro]

Retro: PR #3 (human-authored, closed unmerged)

What happened. rh-hemartin opened PR #3 on 2026-10-26 at 09:52. It moved the fullsend/agents grants out of triagers/writers and into new fullsend-triagers/fullsend-writers teams. That was before fullsend was installed in this repo (PR #4), so no fullsend triage, code, review or fix agent ever ran on it. Only Qodo commented, and it found no issues.

Over the next 4 days, agent PRs #6, #13 and #15 added ederign, jctanner and guyoron1 to triagers. PR #3 had copied the older roster, so merging it would have quietly removed fullsend/agents triage access for those three people. Git showed no textual conflict.

• Triage agent: spotted this semantic conflict correctly each time, on Eder is missing from triagers #5, Add jctanner to triagers #12 and Add Guy to triagers #14, and correctly said there was no conflict on Remove the infrastructure-as-code repository #8 and Add jwm4 as maintainer #19.
• Code agent: carried the warning forward as "whichever merges second" notes in PR bodies.
• Review agent: never mentioned it.

The author closed #3 and asked the agent to redo the change from issue #1. The replacement, agent-authored PR #30, rebuilt the roster from current main and includes all three users. That's a good outcome.

Existing issue (not re-filed). This retro run should not have happened. .fullsend/config.yaml sets retro: enabled: false, but Route reads the config at pull_request.base.sha, and for PR #3 that is 0b54b85, a commit from before the install. The Check agent is enabled step exit 0s when the file is missing, so retro ran anyway. Closes of PR #16 and PR #22 logged retro is disabled as expected. This is fullsend-ai/fullsend#8708 (open), which is about the same stale base SHA bypassing the role filter. This run adds two pieces of evidence:

• A disabled agent (not just a role filter) ran.
• The kill-switch check reads the same stale SHA, so turning the kill switch on in main would not stop dispatch on long-lived PRs.

Proposal. One repo-level proposal: a lint/CI step that prints each user's net access change when peribolos.yaml changes. It would make this kind of silent removal visible to both human reviewers and the review agent. The lint step would build on #25, which was closed without a recorded reason.

Proposals filed

• Lint: print and gate per-user net access changes from peribolos.yaml (in fullsend-ai/org)