--- name: delivery description: Use when committing, pushing, opening or babysitting a PR/MR, merging, or deciding whether verified work is complete under the configured delivery mode. --- # Delivery Use the host's normal Git and forge tools when the user has authorized delivery. Plugin records provide evidence; they do not authorize repository effects or restrict ordinary host tools. For ongoing PR/MR monitoring through merge or an external-intervention blocker, load the retained [babysit PR contract](../../components/babysit-pr/skills/babysit-pr/SKILL.md). When the configured value is unavailable, do not choose a mode. State that `.development-system.toml` is authoritative and summarize all three modes so the user knows what evidence is missing. - A commit must consume a `passing-awaiting-gates-or-review` per-edit checkpoint whose bounded lightweight review has completed. The commit operation itself must trigger the Lefthook pre-commit gate and return both its gate receipt and a signed commit receipt when signing is required. Never run the same gate separately before committing. - A failed pre-commit attempt that created no commit, or a lightweight review requiring remediation, must first record canonical `failing` with actual failure evidence through the public `development-workflow` state machine. Commit and push are prohibited while failing. Preserve staged work; permit only the causal repair followed by fresh tests, review, and a new normal hook-backed commit attempt. Use its recovery helper for a stranded record, never manual checkpoint replacement or fabricated passing receipts. - A tag operation is separate from an increment commit: it consumes the repository-authorized delivered revision and release evidence required by policy, and returns a tag receipt when signing is required. - A push operation must trigger the Lefthook pre-push gate and return its verification plus exact remote readback evidence. Never run that gate separately before pushing. PR/merge operations return their idempotent remote evidence. - Failed pushed CI is blocking recovery work; use the installed Tiber CI-recovery workflow. For a PR/MR check failure, inspect the failed job and step before acting. Classify it as change-caused, transient/flaky, or infrastructure/external; route a causal failure through the configured repair lifecycle, use an authorized rerun for a transient failure, and do not disguise infrastructure failure with an unrelated code edit. Before creating any authorized commit, write a concise Conventional Commit subject and a non-empty body that explains why the change exists. The body must capture the motivation, decision context, tradeoff, or failure being prevented; it must not merely restate the subject or summarize the diff. Reject a subject-only message. Never add `Co-Authored-By` or another AI-attribution trailer. Load the retained `delivery-workflow` contract before the first increment. Strongly recommend the [architecture and decision-history practices](../engineering-standards/references/architecture-policy.md) for project adoption. First inspect the project's current architecture and local policy. Existing conventions and explicit user choices take precedence; installation or setup alone does not adopt these recommendations. Record the accepted subset, any overrides, document paths and prospective adoption boundary in project-owned guidance. Do not add requirements or block delivery solely because an unadopted recommendation is absent. For adopted rules, check the actual candidate and resulting trunk commits, including squash delivery. Use only the corresponding checker rules and retain semantic review evidence separately from mechanical receipts. Offer unadopted practices as recommendations, not findings of noncompliance. PR readiness has independent gates: - Before **any source update to a ready PR/MR**, FIRST return it to draft and confirm that transition succeeded, whether or not auto-merge is enabled. Only then disable auto-merge and confirm it is off before pushing the update. If either prerequisite fails, do not push the update or claim it is safe. - Branch pushes and draft creation may precede local review. Creating a non-draft PR/MR, marking it ready, enabling auto-merge, or entering a merge queue requires a passing current local multi-lens review first. - After source-changing updates, renew affected review before restoring ready status, auto-merge, or queue eligibility. Commit small, coherent passing increments extremely frequently in every delivery mode; do not wait for a feature, vertical slice, or final MVP to be complete. Direct-to-trunk pushes go to origin’s configured trunk; PR/MR pushes go to the selected branch; local-only commits stay local. An explicit no-commit instruction takes precedence and uses the reviewed snapshot route. Each commit runs the configured pre-commit checks at minimum, preserves signing and exact source/message/signature verification, and receives proportional independent review before trunk publication. Review the actual increment and its affected dependencies; unfinished acceptance criteria for later increments are not publication blockers unless the current increment breaks an existing contract. Keep required CI gates and record the exact pushed SHA. Supporting commits are progress toward a slice, not delivered user value. Continue completing usable vertical slices regularly and perform scope-complete final review and acceptance before claiming feature or MVP completion. In direct-to-trunk mode, use one checkout on the configured trunk branch and push directly to origin’s trunk. Do not create a branch, linked worktree, or parallel source checkout unless the user explicitly authorizes a different approach. Serialize mutable work in that checkout, including CI repair; read-only reviewers may inspect it. Use affected independent review for each increment rather than accumulating a completed local batch for one eventual publication review. Keep the immutable task-start baseline for final acceptance while recording each increment’s reviewed scope and exact identity separately. In PR/MR mode, authorized branch pushes and draft creation may precede full local review. Before updating any ready PR/MR, first return it to draft and confirm success, then disable auto-merge and confirm it is off. Stop if either safeguard fails. Current local multi-lens review is required before restoring readiness, enabling auto-merge, or entering the merge queue. Local-only work remains local and uses the authorized commit or no-commit evidence route. In all modes, source, dependency, baseline, scope, or conflict-resolution changes invalidate affected evidence; metadata-only changes require exact identity verification. Required remote CI follows push, and a failure preempts unrelated work through CI recovery. Neither passing CI nor forge approval substitutes for required independent review. Keep checkpoint CI bounded per pushed ref. Use forge/repository coalescing for superseded non-terminal runs, or cancel an obsolete run only when cancellation is authorized; otherwise wait for capacity before another push would create obsolete queued work. Protect the current terminal candidate's exact-SHA run from cancellation, and reject a successful older-SHA run as readiness evidence for the newer checkpoint. If the same-MCP final-review protocol preflight reports a stale skill/runtime or coordinator mismatch, reject terminal delivery, PR/MR creation or update, merge, and readiness claims until the required recovery succeeds. That mismatch does not retroactively forbid a separately authorized intermediate or terminal-finding-remediation checkpoint: it may still preserve a local commit or push a draft branch; trunk push and PR readiness remain blocked, but it cannot be called terminal delivery or readiness evidence. Always state each protected action explicitly: never infer permission to force-push, merge, or delete remote state. When trunk CI failure interrupts WIP, load the [trunk recovery protocol](../../components/development-discipline/skills/ci-failure-follow-up/references/trunk-wip-recovery.md) for exact stash identity, unpublished commit handling, index-preserving restoration, conflicts and interruption recovery. Keep the single checkout and incident-owner boundaries while preserving WIP.