Split out of PR #67, where CodeRabbit raised it as a Major (Security & Privacy) finding. It was declined there as out of scope: #67's diff on qa/skills/headless/SKILL.md is frontmatter-only (a single description field, body verified byte-identical to main), so the behaviour predates that change and is untouched by it.
The finding
qa:headless Phase 8 edits files and runs git commit with no approval step. CodeRabbit's recommendation:
Report each finding and proposed diff first. Require explicit approval before editing files and before running git commit.
SkillSpector flagged the same shape as [EA2] Autonomous Decision Making: a skill enabling high-impact autonomous operations without human-in-the-loop verification.
Why it deserves its own PR
This is a real behavioural change to how the skill runs, not a docs fix. It needs its own QA plan and review on its own terms: today qa:headless is expected to find bugs, fix them with atomic commits, and re-verify in one unattended pass, and that autonomy is part of its value. Adding a gate changes the contract for anyone who relies on it. Worth deciding deliberately rather than as a drive-by.
Also noted
The same file carries 89 U+2014 em-dash characters in its body, which violates the standing no-em-dash rule. Mechanical to fix, but it touches ~89 lines of a 66KB body, so it does not belong in a metadata PR either. Reasonable to fold into the same PR as the approval gate, or do as its own sweep.
Suggested scope
Split out of PR #67, where CodeRabbit raised it as a Major (Security & Privacy) finding. It was declined there as out of scope: #67's diff on
qa/skills/headless/SKILL.mdis frontmatter-only (a singledescriptionfield, body verified byte-identical tomain), so the behaviour predates that change and is untouched by it.The finding
qa:headlessPhase 8 edits files and runsgit commitwith no approval step. CodeRabbit's recommendation:SkillSpector flagged the same shape as
[EA2] Autonomous Decision Making: a skill enabling high-impact autonomous operations without human-in-the-loop verification.Why it deserves its own PR
This is a real behavioural change to how the skill runs, not a docs fix. It needs its own QA plan and review on its own terms: today
qa:headlessis expected to find bugs, fix them with atomic commits, and re-verify in one unattended pass, and that autonomy is part of its value. Adding a gate changes the contract for anyone who relies on it. Worth deciding deliberately rather than as a drive-by.Also noted
The same file carries 89 U+2014 em-dash characters in its body, which violates the standing no-em-dash rule. Mechanical to fix, but it touches ~89 lines of a 66KB body, so it does not belong in a metadata PR either. Reasonable to fold into the same PR as the approval gate, or do as its own sweep.
Suggested scope