Review agent intent-coherence sub-agent should question changes motivated by external scoring frameworks

[fullsend-ai-retro]

What happened

On PR #3415, the PR description stated the goal was 'improving the repo's AgentReady score to Gold certification.' The PR bundled 8+ independent changes (gitignore, commitlint, depguard, Claude config, PR template, docs) under this umbrella. The commitlint commit message (b8dd6d47) explained: 'The existing check-commit-message hook validates format but is not recognized by standard conventional-commit tooling.' The final removal commit (ebe9a83) noted: 'The agentready conventional_commits assessor only checks for the config file' — revealing the motivation was passing an assessment checklist, not team need.

What could go better

The review agent treated the 'AgentReady Gold certification' framing as informational context but did not use it to assess whether individual changes were justified on their own merits. When a PR bundles multiple independent changes under an external compliance/scoring umbrella, the agent should evaluate each change independently. The commitlint addition was technically correct but driven by a scoring framework rather than genuine team need. This is a medium-confidence finding — the signal was present in the PR description but requires nuanced judgment about when external scoring motivations should trigger additional scrutiny.

Proposed change

Add guidance to the intent-coherence sub-agent prompt: when a PR description references external compliance frameworks, certification programs, or scoring systems as the motivation for changes, evaluate each individual change on its own merits rather than accepting the umbrella justification. If a change appears to exist primarily to satisfy an external checklist rather than address a genuine team need, flag it as a low-severity finding noting the observation and suggesting the author confirm team alignment. This is a soft signal, not a blocking finding — the goal is to prompt the conversation that human reviewer st3penta naturally initiated.

Validation criteria

On future PRs that bundle multiple changes under an external compliance or scoring framework, the review agent produces a low-severity finding noting that individual changes should be evaluated on their own merits and prompting confirmation of team alignment for any that appear checkbox-driven.

---

Generated by retro agent from conforma/cli#3415

[fullsend-ai-prioritize]

RICE Priority Score: 0.7

Score breakdown

Dimension	Score	Reasoning
Reach	1	This affects all users of the review agent's intent-coherence sub-agent, but the specific scenario it addresses — PRs bundling changes under external compliance or scoring frameworks — is relatively uncommon. Scoring as a moderate number of users who would encounter this pattern.
Impact	0.5	Low impact per affected user. The proposed change adds a soft, low-severity signal rather than preventing a critical failure. The original PR review already showed a human reviewer (st3penta) naturally caught the issue, so a workaround exists. This is a quality-of-life improvement to agent review thoroughness.
Confidence	0.7	The issue is well-described with a concrete retroactive example, a specific proposed prompt change, and clear validation criteria. However, it was auto-generated by a retro agent rather than reported by a user experiencing pain, which slightly reduces confidence in the real-world frequency and importance of this scenario.
Effort	0.5	The proposed change is scoped to adding guidance text to the intent-coherence sub-agent prompt — essentially a prompt engineering change. This is a small, well-defined modification, though it requires thoughtful wording and validation against test cases to avoid false positives.

Formula: (1 x 0.5 x 0.7) / 0.5 = 0.7