review-postinglisted
Install: claude install-skill m4ttstack/skills
# Review posting (execution only)
Turning a decided review selection into what lands on the MR/PR: which
findings get a thread, what disposition closes it, how the summary reads,
how the close ends. This part owns execution. The decision -- which
severity levels post and what disposition closes the review -- belongs to
the caller.
## Caller inputs
- A decided selection: `{levels: [...], disposition: "comment" | "approve"
| "request_changes"}`. `levels` names the severity buckets to post
(whichever of Critical / Important / Minor the draft carries); `disposition`
names the one the caller already chose. Arriving without both is a caller
bug -- see the guard below.
- The draft, in the review flow's Strengths / Issues shape: Strengths /
Issues (Critical / Important / Minor, each `file:line`) / Assessment
(yes | no | with fixes), when it is in context -- take it as given, never
re-derive or re-judge a finding here. When the draft is not in context
(the parked-resume case), read it from the written report file instead;
the report's fixed severity buckets are enough to execute from.
- A postable target: an MR/PR whose posting mechanics -- anchoring an inline
comment to a line, verifying it landed, composing the summary body -- the
caller owns. This part decides how to execute, not what posts or where.
- **Callers:** the review verb's Deliver step, a domain adapter's
teammate-review skill (the harvest `review` skill, after its thinning
pass), or any sessi