Part of #1311. Priority: after P0. Needs a maintainer decision before any of the 23 issues is worked.
Why
backlog-burndown merged in #1291 after sixteen review rounds with a known defect list, and now has 23 open issues against it (#1272 to #1310, minus the ones against sibling skills), filed within two hours of its merge. It is 37.7 KB, the largest skill in the tree, and it restates rules from AGENTS.md, drive-pr, and pr-review-conduct (#1305). Its central rule is unsatisfiable in this repository until the ledger sub-issue lands (#1290). Working its 23 issues one at a time, each a whole-unit pass over a 37 KB unit, is the pump at its worst. Nothing carries it yet, so nothing downstream depends on its current text.
The decision
Pick one:
- Hold it experimental. Mark the skill experimental in its frontmatter and in
AGENTS.md's map row, forbid dispatched runs until the P0 sub-issues land and the ledger stops colliding, and work its 23 issues as one rewrite under the normative format rather than 23 pull requests.
- Shrink it. Cut it to the round loop, ranking, file claims, and the promotion boundary. Everything it says about how a worker drives a pull request, disposes of findings, or escalates is
drive-pr's and pr-review-conduct's and becomes a pointer. Most of the 23 issues are against the parts that would be cut.
- Keep it as is and work the 23 issues through the normal loop, accepting the cost.
The recommendation is 2, done as one rewrite after the skill-interface sub-issue defines what a skill declares, since the orchestrator is the skill that most needs its inputs, outputs, and tracked writes stated.
Scope, once decided
Rules that bind this task
- No dispatched run of the skill until the decision is recorded and the ledger collision is fixed.
- The skill owns orchestration. It does not own how a worker drives a pull request.
Acceptance
- The decision is recorded on this issue.
- The 23 issues are closed or consolidated into one tracker, with each one's disposition named.
Relates #1272, #1273, #1274, #1275, #1276, #1277, #1278, #1279, #1280, #1281, #1285, #1286, #1289, #1290, #1293, #1294, #1301, #1302, #1303, #1305, #1306, #1307, #1308, #1309, #1310.
Part of #1311. Priority: after P0. Needs a maintainer decision before any of the 23 issues is worked.
Why
backlog-burndownmerged in #1291 after sixteen review rounds with a known defect list, and now has 23 open issues against it (#1272 to #1310, minus the ones against sibling skills), filed within two hours of its merge. It is 37.7 KB, the largest skill in the tree, and it restates rules fromAGENTS.md,drive-pr, andpr-review-conduct(#1305). Its central rule is unsatisfiable in this repository until the ledger sub-issue lands (#1290). Working its 23 issues one at a time, each a whole-unit pass over a 37 KB unit, is the pump at its worst. Nothing carries it yet, so nothing downstream depends on its current text.The decision
Pick one:
AGENTS.md's map row, forbid dispatched runs until the P0 sub-issues land and the ledger stops colliding, and work its 23 issues as one rewrite under the normative format rather than 23 pull requests.drive-pr's andpr-review-conduct's and becomes a pointer. Most of the 23 issues are against the parts that would be cut.The recommendation is 2, done as one rewrite after the skill-interface sub-issue defines what a skill declares, since the orchestrator is the skill that most needs its inputs, outputs, and tracked writes stated.
Scope, once decided
AGENTS.md's map row and Session Scope exception updated to match, per AGENTS.md: the orchestrating-session exception leaves that session with no ending condition #1289 and AGENTS.md Session Scope has no shape for a drive-pr to-main session #1286.Rules that bind this task
Acceptance
Relates #1272, #1273, #1274, #1275, #1276, #1277, #1278, #1279, #1280, #1281, #1285, #1286, #1289, #1290, #1293, #1294, #1301, #1302, #1303, #1305, #1306, #1307, #1308, #1309, #1310.