AGENTS.md: add reverse-direction docs-to-code verification guidance

[fullsend-ai-retro]

What happened

PR #6281 removed mint enrollment guidance from getting-started docs, claiming the hosted mint was now public and enrollment was no longer needed. The review agent (run 1, run 2) approved both commits without checking the underlying code. Human reviewer waynesun09 discovered that DefaultMintURL in internal/config/defaults.go and internal/cli/admin.go still pointed to the legacy Cloud Run URL — a prior code fix (PR #6261) had been silently reverted by merge PR #6257. The PR was closed and replaced by PR #6290.

AGENTS.md currently instructs: "When removing/renaming/changing CLI commands or user-facing features, grep docs/ for references and update them." This covers code-to-docs direction but not the reverse. Neither the human author nor the review agent verified that the code actually supported the docs' new behavioral claims before the human review.

What could go better

AGENTS.md has forward-direction guidance (code changes should update docs) but no reverse-direction guidance (docs behavioral changes should verify code). Adding reverse-direction guidance would help both human contributors and the review agent catch cases where docs make claims the code doesn't support.

Confidence is high that this gap contributed to the premature approval — the review agent treated this as a pure docs change and never read the Go source files. Confidence is moderate that adding the guidance alone would change outcomes — the review agent would also need to act on it by reading relevant code files, which depends on agent-layer capabilities tracked in agents#265 and fullsend#1542. However, the guidance would immediately help human contributors and provides a foundation for agent improvements.

Proposed change

Add a bullet to the documentation-maintenance section of AGENTS.md, near the existing "When removing/renaming/changing CLI commands" guidance:

When docs remove, weaken, or change behavioral guidance about CLI commands, enrollment workflows, default URLs, or system requirements (e.g., removing a prerequisite step, claiming a feature is now automatic), verify that the underlying code (constants, defaults, validation logic) actually implements the claimed behavior. Check the relevant source files — not just the docs diff.

This complements the existing forward-direction guidance and gives both agents and humans a concrete instruction to cross-reference code when docs behavioral claims change.

Validation criteria

The next 5 docs-only PRs that change behavioral guidance about CLI features should show evidence of reverse-direction code verification — either in the PR description, review comments, or commit messages. At least one should reference this AGENTS.md guidance as the prompt for verification.

---

Generated by retro agent from #6281

[fullsend-ai-prioritize]

RICE Priority Score: 3.2

Score breakdown

Dimension	Score	Reasoning
Reach	1	This affects all contributors (human and AI agents) who submit or review docs PRs that change behavioral guidance. That's a moderate number of people involved in the fullsend development workflow, but not end-users directly.
Impact	1	Adding reverse-direction guidance to AGENTS.md is a noticeable process improvement that could prevent docs-code mismatches like the one in PR #6281. However, the issue itself notes only moderate confidence that guidance alone changes outcomes — agent capabilities tracked elsewhere are also needed. Immediate benefit is primarily for human contributors.
Confidence	0.8	High confidence. The issue is exceptionally well-described with clear context (the specific PR that motivated it), a precise proposed change (a single bullet in AGENTS.md), and concrete validation criteria. The scope is unambiguous.
Effort	0.25	Trivial — this is adding a single bullet point to an existing section of AGENTS.md. The proposed wording is already drafted in the issue body. This is essentially a one-liner documentation change.

Formula: (1 x 1 x 0.8) / 0.25 = 3.2