docs(adr): add ADR 0065 for human-gated permission adjustments

[fullsend-ai-coder]

Records the three-layer design (CLI policy, mint mechanics, automation privileges) and threat model for label-gated elevated agent permissions. Updates architecture.md to reflect the decision.

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

---

Closes #2821

Post-script verification

• Branch is not main/master (agent/2821-adr-human-gated-permissions)
• Secret scan passed (gitleaks — 60e4a9e16f16157fdf61d26a2c73138192befe68..HEAD)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[github-actions]

Site preview

Preview: https://705de61a-site.fullsend-ai.workers.dev

Commit: 9b898c3640a59cc43b852aa0cadb38a559b865e7

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[rh-hemartin]

@ifireball should be you assigned to this?

[ifireball]

@ifireball should be you assigned to this?

Probably, I wasn't expecting to see code run automatically on an ADR issue...

[rh-hemartin]

Feel free to close it if it does not make sense.

[ifireball]

Closing, #5311 defines the ADR we need better, might re-open or take another shot at this ADR if I think its needed for the specific privilege escalation we need in "code" to handle workflow writes.

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 1:19 PM UTC · Completed 1:30 PM UTC
Commit: 9b898c3 · View workflow run →

[fullsend-ai-retro]

Retro: PR #2830 — ADR 0065 for human-gated permission adjustments

Outcome: PR closed without merging after 18 days. The code agent's work was wasted.

Timeline

1. Jul 1 07:12 — @ifireball filed #2821 requesting an ADR to record design decisions for human-gated permission adjustments.
2. Jul 1 07:12–07:18 — Triage agent ran, categorized issue as documentation, action sufficient → ready-to-code applied automatically by post-triage.
3. Jul 1 07:18–07:28 — Code agent ran ($1.69, ~5 min active), wrote ADR 0065 with status "Accepted", updated docs/architecture.md, created PR docs(adr): add ADR 0065 for human-gated permission adjustments #2830.
4. No review agent ran — the review agent infrastructure was not deployed until ~July 15.
5. Jul 8 — @rh-hemartin flagged the PR for @ifireball. @ifireball responded: "I wasn't expecting to see code run automatically on an ADR issue."
6. Jul 19 — @ifireball closed the PR, noting that #5311 (a more general ADR for named mint privilege levels) should land first.

Key findings

• Root cause of waste: The post-triage script treats documentation as a low-risk category and auto-promotes all documentation issues to ready-to-code. ADR issues are fundamentally different from routine docs — they require team discussion and consensus before an agent writes them.
• ADR status gap: The code agent wrote the ADR with "Accepted" status for a design that hadn't been agreed upon. This is already tracked in #5239.
• Agent execution quality was good — the code agent did thorough research, smart ADR number collision avoidance, and comprehensive verification. The problem was task routing, not execution.

Proposals filed

1. Post-triage should route ADR issues to human prioritization instead of auto-promoting to ready-to-code

Proposals filed

• Post-triage should route ADR issues to human prioritization, not auto-promote to ready-to-code (in fullsend-ai/fullsend)