reconcile-build
Skill16 starsChanged 4 months agoMITView on GitHub
The closeout procedures the tail REC.* tickets invoke — backlog + operational readiness, spec reconciliation, and the integration plan — for a multi-session build. Use it when a build's tickets have landed and the tail rows (from decompose-spec) call 'invoke reconcile-build mode=…', the way P22.x tickets called refresh-repo-docs. Reads the committed build memory (DEFERRALS, ADRs, OPEN FINDINGS, the coverage matrix, BUILD_INDEX) and writes the backlog / reconciliation / integration artifacts. Not a sequencer and not a merge tool.
In this repository10 other files it gives its agents.
- address-pr-commentsSkill
- agent-docsSkill
- build-memorySkill
- decompose-specSkill
- implement-specSkill
- orchestrate-buildSkill
- refresh-repo-docsSkill
- review-prSkill
- self-reviewSkill
- synthesize-specSkill
The file
Files
Reconcile Build
Hold the closeout procedures a multi-session build runs after its tickets land — so they live in one place
with their templates and checkers, and the tail REC.* tickets just say "invoke reconcile-build mode=…"
(the way the docs tickets say "invoke refresh-repo-docs"). Folding these into orchestrate-build would push
that SKILL past the line ceiling and mix sequencing with content; this skill is the content.
Precedent.
agent-docs/refresh-repo-docsown the docs procedures theDOC.*tickets invoke;reconcile-buildowns the reconciliation procedures theREC.*tickets invoke. It reads and writes the committed build memory defined inskills/build-memory/layout.md(it cites that file, does not restate it):DEFERRALS.md,docs/adr/, the ledger'sOPEN FINDINGS,COVERAGE_MATRIX.csv,BUILD_INDEX.md, and theBACKLOG.*/*_PLAN.md/OPERATIONAL_READINESS.mdoutputs underdocs/build/.
Orient
Resolve the memory root (build-memory memory-root.sh); confirm docs/build/ exists and the validator is
green (check-build-memory.sh .) before reconciling — reconciliation over an inconsistent layout is noise. Then
follow the mode.
The modes
| Mode | File | Produces |
backlog | modes/backlog.md | docs/build/BACKLOG.csv, BACKLOG.md, OPERATIONAL_READINESS.md — every owed thing in exactly one row; readiness with a proof on every critical-path step. |
spec | modes/spec.md | docs/build/TICKET_VS_SPEC.md, SPEC_RECONCILIATION_PLAN.md — what the build did vs what the spec said, and the amendments to fold back. |
integration | modes/integration.md | docs/build/INTEGRATION_PLAN.md (the PR graph with its check states), a release-notes draft, and a planning/<date -u +%F>_decision-memo.md seeding the next round. |
The one discipline that matters
Every owed thing lands in exactly one backlog row. The failure mode of build closeout is debt that is
"tracked" in five places and therefore nowhere. backlog gathers every source — each OPEN/PARTIAL deferral,
each ADR revisit trigger, each OPEN FINDINGS entry, each spec-mandated register's deferred row, each non-MET
coverage-matrix row — into a single sources cell, and scripts/check-backlog.sh fails if any source appears
zero times or more than once. That is the backlog's contract: complete and non-duplicating.
Scripts (bash 3.2+, read-only)
| Script | Role |
scripts/check-backlog.sh | verify the backlog is complete + non-duplicating with live homes; verdicts consistent with DEFERRALS (BM-VERDICT-01); the two sums vs CAP.3; ids unique; statuses valid |
scripts/merge-dryrun.sh | report, per chain PR/branch, whether a clean merge is possible and (--ci) its current check state — never merges |
What this skill does NOT do
- No sequencing —
orchestrate-buildruns theREC.*tickets; this skill is what they invoke. - No merging, no CI fixing, no release — CI state is read (read-only) and reported;
merge-dryrun.shis read-only; the actual merge/release is the operator's, from the copy-pasteable procedureintegrationwrites. - No silent spec edits —
specwrites proposals; it applies an amendment only where the plan is ticked andapply_amendments=true, always throughspec_srcwhen it exists, with a manifest amendment line + an ADR.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

