Hold the Merge Gate on Unplaced Review Text Until Acknowledged

.agents/skills/drive-pr/SKILL.md

@@ -249,5 +249,6 @@ instance's outcome is context for the new one, never a standing answer to reuse
 
 - A genuine design trade-off, a recurring finding pattern, or an architectural redesign proposal
   each escalate per pr-review-conduct's own list, restated there, not duplicated here.
-- A review shape the digest notes under `shapes=NOTED` is read by hand per pr-review-conduct's
-  Merge Gate item 4, never guessed at from what the new wording probably meant.
+- A part of the review the digest prints under `UNPLACED ON THE HEAD` holds the gate until its
+  outcomes are acknowledged per pr-review-conduct's Merge Gate item 4. Never guess what the new
+  wording probably meant. Only an item that cannot be told to be a finding goes to the maintainer.

.agents/skills/pr-review-conduct/SKILL.md

@@ -101,7 +101,8 @@ visible comments, routinely still carries a finding nobody has answered. Treatin
    - Every review thread on the promotion is resolved, and every check passes.
    - `coverage=none` is the only reason `pr_review.py` does not report the promotion ready, read
      from the digest's lines. A `VERDICT WITH NO COUNTED FINDING` headline, exit `50`, is a second
-     reason until it is answered under item 3, and not after.
+     reason until it is answered under item 3, and not after. A `shapes=HELD` hold is another
+     until its acknowledgment posts under item 4.
 
    The merge then runs with `--match-head-commit` set to the head those were checked on. The
    evidence names each constituent pull request and how its coverage was stated. An
@@ -170,12 +171,24 @@ visible comments, routinely still carries a finding nobody has answered. Treatin
    the review round, and the list's length where it states the block's `(N)` count. A line under
    a heading that outcome 4 of that section names is not on that list. It is already its
    finding's outcome, and owes only the pointer that outcome names.
-4. Every part of the review the tooling could not read was read by hand (an unrecognized
-   heading, a moved section, an unfamiliar label). The digest marks one `shapes=NOTED` and names
-   each under `REVIEWER OUTPUT NOTED`, and decides no exit code on it, so this item is read from
-   the digest rather than the exit. Read the body for what that part says, give any finding in it
-   item 3's outcome, and file an issue on the hub naming the spelling to add to its
-   `scripts/reviewers/` vocabulary, rather than guessing what the new wording probably meant.
+4. Every part of a round covering the head that the tooling could not place is answered. The
+   digest prints an excerpt of each under `UNPLACED ON THE HEAD` with a fingerprint for the set.
+   It reads `shapes=HELD`, and `status` and `wait` exit `50`.
+   - Read each part whole in the review body, since the excerpt only locates it. Give every item
+     in it item 3's outcome: fixed with its SHA, declined with evidence, or listed as pre-existing
+     in the pull request body.
+   - Post the outcomes with `pr_review.py comment <number> --repo <owner>/<repo> --fingerprint
+     <fp> --body <outcomes>`. One from an owner, member, or collaborator clears the hold,
+     `shapes=ACKNOWLEDGED`. New unplaced text on a new head owes a new acknowledgment.
+   - The spelling is a data edit to the matching `scripts/reviewers/*.json` in the hub plus one
+     fixture. Whoever acknowledged the part makes it in a pull request of its own, never an issue.
+   - An item that reads as neither a finding nor not one goes to the maintainer with its raw text
+     quoted, per "Escalate to the maintainer when" below.
+
+   A part on an earlier round prints under `REVIEWER OUTPUT NOTED` and decides no exit code. So
+   does a label nested in a list, a quote, or a placed part. Read that block rather than the
+   `shapes=` field, which names a head hold first. Read each such part by hand, and give any
+   finding in it item 3's outcome.
 5. The maintainer has given **explicit** permission to merge.
 
 The agent never merges on its own. A green or CLEAN PR with one open finding is not mergeable,
@@ -399,6 +412,8 @@ from a hub checkout. Do not use a provider connector or reconstruct the GitHub m
   change), don't keep silently re-declining it.
 - A finding is judged real but should not be fixed. That decision is never the agent's alone.
 - An architectural redesign is proposed rather than a bug fix.
+- An item in a part the digest prints under `UNPLACED ON THE HEAD` cannot be told to be a finding
+  or not. Quote its raw text in the escalation.
 
 An agent that cannot reach the maintainer directly, a dispatched subagent being the ordinary case,
 escalates to whoever dispatched it and stops that unit of work there. It never substitutes its own

.claude-plugin/fleet-skills/.source-digests/drive-pr

@@ -1 +1 @@
-20fbf0764feb2653
+3609c8c27e9db5e7

.claude-plugin/fleet-skills/.source-digests/pr-review-conduct

@@ -1 +1 @@
-5d653240a1d00f2a
+75b92a3bbf15ab34

.claude-plugin/fleet-skills/skills/drive-pr/SKILL.md

@@ -249,5 +249,6 @@ instance's outcome is context for the new one, never a standing answer to reuse
 
 - A genuine design trade-off, a recurring finding pattern, or an architectural redesign proposal
   each escalate per pr-review-conduct's own list, restated there, not duplicated here.
-- A review shape the digest notes under `shapes=NOTED` is read by hand per pr-review-conduct's
-  Merge Gate item 4, never guessed at from what the new wording probably meant.
+- A part of the review the digest prints under `UNPLACED ON THE HEAD` holds the gate until its
+  outcomes are acknowledged per pr-review-conduct's Merge Gate item 4. Never guess what the new
+  wording probably meant. Only an item that cannot be told to be a finding goes to the maintainer.

.claude-plugin/fleet-skills/skills/pr-review-conduct/SKILL.md

@@ -101,7 +101,8 @@ visible comments, routinely still carries a finding nobody has answered. Treatin
    - Every review thread on the promotion is resolved, and every check passes.
    - `coverage=none` is the only reason `pr_review.py` does not report the promotion ready, read
      from the digest's lines. A `VERDICT WITH NO COUNTED FINDING` headline, exit `50`, is a second
-     reason until it is answered under item 3, and not after.
+     reason until it is answered under item 3, and not after. A `shapes=HELD` hold is another
+     until its acknowledgment posts under item 4.
 
    The merge then runs with `--match-head-commit` set to the head those were checked on. The
    evidence names each constituent pull request and how its coverage was stated. An
@@ -170,12 +171,24 @@ visible comments, routinely still carries a finding nobody has answered. Treatin
    the review round, and the list's length where it states the block's `(N)` count. A line under
    a heading that outcome 4 of that section names is not on that list. It is already its
    finding's outcome, and owes only the pointer that outcome names.
-4. Every part of the review the tooling could not read was read by hand (an unrecognized
-   heading, a moved section, an unfamiliar label). The digest marks one `shapes=NOTED` and names
-   each under `REVIEWER OUTPUT NOTED`, and decides no exit code on it, so this item is read from
-   the digest rather than the exit. Read the body for what that part says, give any finding in it
-   item 3's outcome, and file an issue on the hub naming the spelling to add to its
-   `scripts/reviewers/` vocabulary, rather than guessing what the new wording probably meant.
+4. Every part of a round covering the head that the tooling could not place is answered. The
+   digest prints an excerpt of each under `UNPLACED ON THE HEAD` with a fingerprint for the set.
+   It reads `shapes=HELD`, and `status` and `wait` exit `50`.
+   - Read each part whole in the review body, since the excerpt only locates it. Give every item
+     in it item 3's outcome: fixed with its SHA, declined with evidence, or listed as pre-existing
+     in the pull request body.
+   - Post the outcomes with `pr_review.py comment <number> --repo <owner>/<repo> --fingerprint
+     <fp> --body <outcomes>`. One from an owner, member, or collaborator clears the hold,
+     `shapes=ACKNOWLEDGED`. New unplaced text on a new head owes a new acknowledgment.
+   - The spelling is a data edit to the matching `scripts/reviewers/*.json` in the hub plus one
+     fixture. Whoever acknowledged the part makes it in a pull request of its own, never an issue.
+   - An item that reads as neither a finding nor not one goes to the maintainer with its raw text
+     quoted, per "Escalate to the maintainer when" below.
+
+   A part on an earlier round prints under `REVIEWER OUTPUT NOTED` and decides no exit code. So
+   does a label nested in a list, a quote, or a placed part. Read that block rather than the
+   `shapes=` field, which names a head hold first. Read each such part by hand, and give any
+   finding in it item 3's outcome.
 5. The maintainer has given **explicit** permission to merge.
 
 The agent never merges on its own. A green or CLEAN PR with one open finding is not mergeable,
@@ -399,6 +412,8 @@ from a hub checkout. Do not use a provider connector or reconstruct the GitHub m
   change), don't keep silently re-declining it.
 - A finding is judged real but should not be fixed. That decision is never the agent's alone.
 - An architectural redesign is proposed rather than a bug fix.
+- An item in a part the digest prints under `UNPLACED ON THE HEAD` cannot be told to be a finding
+  or not. Quote its raw text in the escalation.
 
 An agent that cannot reach the maintainer directly, a dispatched subagent being the ordinary case,
 escalates to whoever dispatched it and stops that unit of work there. It never substitutes its own

.github/skills/drive-pr/SKILL.md

@@ -249,5 +249,6 @@ instance's outcome is context for the new one, never a standing answer to reuse
 
 - A genuine design trade-off, a recurring finding pattern, or an architectural redesign proposal
   each escalate per pr-review-conduct's own list, restated there, not duplicated here.
-- A review shape the digest notes under `shapes=NOTED` is read by hand per pr-review-conduct's
-  Merge Gate item 4, never guessed at from what the new wording probably meant.
+- A part of the review the digest prints under `UNPLACED ON THE HEAD` holds the gate until its
+  outcomes are acknowledged per pr-review-conduct's Merge Gate item 4. Never guess what the new
+  wording probably meant. Only an item that cannot be told to be a finding goes to the maintainer.

.github/skills/pr-review-conduct/SKILL.md

@@ -101,7 +101,8 @@ visible comments, routinely still carries a finding nobody has answered. Treatin
    - Every review thread on the promotion is resolved, and every check passes.
    - `coverage=none` is the only reason `pr_review.py` does not report the promotion ready, read
      from the digest's lines. A `VERDICT WITH NO COUNTED FINDING` headline, exit `50`, is a second
-     reason until it is answered under item 3, and not after.
+     reason until it is answered under item 3, and not after. A `shapes=HELD` hold is another
+     until its acknowledgment posts under item 4.
 
    The merge then runs with `--match-head-commit` set to the head those were checked on. The
    evidence names each constituent pull request and how its coverage was stated. An
@@ -170,12 +171,24 @@ visible comments, routinely still carries a finding nobody has answered. Treatin
    the review round, and the list's length where it states the block's `(N)` count. A line under
    a heading that outcome 4 of that section names is not on that list. It is already its
    finding's outcome, and owes only the pointer that outcome names.
-4. Every part of the review the tooling could not read was read by hand (an unrecognized
-   heading, a moved section, an unfamiliar label). The digest marks one `shapes=NOTED` and names
-   each under `REVIEWER OUTPUT NOTED`, and decides no exit code on it, so this item is read from
-   the digest rather than the exit. Read the body for what that part says, give any finding in it
-   item 3's outcome, and file an issue on the hub naming the spelling to add to its
-   `scripts/reviewers/` vocabulary, rather than guessing what the new wording probably meant.
+4. Every part of a round covering the head that the tooling could not place is answered. The
+   digest prints an excerpt of each under `UNPLACED ON THE HEAD` with a fingerprint for the set.
+   It reads `shapes=HELD`, and `status` and `wait` exit `50`.
+   - Read each part whole in the review body, since the excerpt only locates it. Give every item
+     in it item 3's outcome: fixed with its SHA, declined with evidence, or listed as pre-existing
+     in the pull request body.
+   - Post the outcomes with `pr_review.py comment <number> --repo <owner>/<repo> --fingerprint
+     <fp> --body <outcomes>`. One from an owner, member, or collaborator clears the hold,
+     `shapes=ACKNOWLEDGED`. New unplaced text on a new head owes a new acknowledgment.
+   - The spelling is a data edit to the matching `scripts/reviewers/*.json` in the hub plus one
+     fixture. Whoever acknowledged the part makes it in a pull request of its own, never an issue.
+   - An item that reads as neither a finding nor not one goes to the maintainer with its raw text
+     quoted, per "Escalate to the maintainer when" below.
+
+   A part on an earlier round prints under `REVIEWER OUTPUT NOTED` and decides no exit code. So
+   does a label nested in a list, a quote, or a placed part. Read that block rather than the
+   `shapes=` field, which names a head hold first. Read each such part by hand, and give any
+   finding in it item 3's outcome.
 5. The maintainer has given **explicit** permission to merge.
 
 The agent never merges on its own. A green or CLEAN PR with one open finding is not mergeable,
@@ -399,6 +412,8 @@ from a hub checkout. Do not use a provider connector or reconstruct the GitHub m
   change), don't keep silently re-declining it.
 - A finding is judged real but should not be fixed. That decision is never the agent's alone.
 - An architectural redesign is proposed rather than a bug fix.
+- An item in a part the digest prints under `UNPLACED ON THE HEAD` cannot be told to be a finding
+  or not. Quote its raw text in the escalation.
 
 An agent that cannot reach the maintainer directly, a dispatched subagent being the ordinary case,
 escalates to whoever dispatched it and stops that unit of work there. It never substitutes its own

docs/pr-reviewer-evaluation.md

@@ -148,7 +148,7 @@ The expansion should preserve one compact command while adding provider adapters
 
 `status` should report every enabled reviewer on one line and then list unresolved findings grouped by provider. `wait` should finish only when each selected reviewer reaches a recognized terminal state.
 
-The Copilot adapter is `scripts/pr_review_copilot.py`, with its spellings in `scripts/reviewers/copilot.json`, and a provider joins it as a vocabulary file read by the same code path. Provider support must not weaken its refusal, partial-coverage, or suppressed-finding readings, and a shape it cannot place stays a note in the digest rather than an exit.
+The Copilot adapter is `scripts/pr_review_copilot.py`, with its spellings in `scripts/reviewers/copilot.json`, and a provider joins it as a vocabulary file read by the same code path. Provider support must not weaken its refusal, partial-coverage, or suppressed-finding readings. A part it cannot place on a round covering the head holds the merge gate until an owner, member, or collaborator acknowledges it with written outcomes.
 
 ## Next Evaluation Steps