kickoff-staging-placement — dispatch-input placement

Census id: F17 · Family: F — rules corpus + generated-rule tooling + shipped audit scripts · Provenance: framework pin b069c59328aa8e08671d3a19fd25efa618dfc1c5 (staging FETCH_HEAD, read 2026-09-11; re-anchored from census pin a1337cb301). Every file:line below is at that pin. Raw reference drafted by stage E4 — factual, unpolished by design.

What it bans

Dispatching against a kickoff that only exists on a feature branch, and authoring mistakes on this surface that «render as success». .claude/rules/kickoff-staging-placement.md:7 «# Kickoff staging-placement — dispatch-input discipline»; the rule: :21 «Kickoffs under .claude/orchestrator-prompts/<umbrella>/kickoff.md are tracked files, read from the staging branch by every dispatch consumer:» and :26 «A kickoff that exists only on a feature worktree branch is invisible to them.» The binding sequence: :28 «Sequence (binding): author the kickoff → merge it to staging (PR, squash) → only then hand out /pipeline <umbrella> or initiate an aif dispatch.» Origin: :17 «The first time, a /pipeline pipeline-i18n-fix run on staging resolved to a stale same-named plan and fixed the wrong thing (PR #578 ≠ the intended output-directive).» Class header: :12 «> Class: A — companion principle test shipped at [packages/core/principles/44-kickoff-authoring-traps.test.ts] (2026-09-02), plus two earlier-channel gates … §1 itself stays ungated — merge timing … is not assertable at any branch-scoped channel».

Never (fires)

The three anti-patterns («all three share one shape: an authoring mistake on this surface that renders as success», :53): .claude/rules/kickoff-staging-placement.md:55 «### §5.1 #dispatch-before-staging» — «The dispatch session (on staging) silently can't find it and either no-ops or resolves to a stale same-named artifact.»; :59 «### §5.2 #canon-symlink-swallows-commit» — «…the commit takes a symlink instead of the content» (:61), measured: :65 «commit 9e046c6d55 carried 120000 blob 2d02… while the 232-line kickoff existed only in $CANON»; :71 «### §5.3 #kickoff-name-near-miss» — «Naming a stage kickoff so that it just misses the stage-kickoff family, making every gate treat it as a deliberate sidecar and report green having examined nothing.»

Always (clean)

Write + add in one step, and a resolvable name: :67 «Counter: write the file and git add it in one step.» and :79 «Counter: a kickoff-* name must resolve to one of exactly two things — the stage form kickoff-<letter><digit>[alnum].md, or a named sidecar (kickoff[-<stage>].<kind>.md, or the exact-name allowlist). Anything else is a loud error naming both alternatives, never a silent reclassification.»

Enforcement channels

• Rule-index row: .claude/rules/00-rule-index.md:25 «| kickoff-staging-placement.md | A | editing/creating any file under .claude/orchestrator-prompts/<umbrella>/. | paths:(1), edit-time inject |»
• Edit-time inject, 1 glob: .claude/rules/kickoff-staging-placement.md:9 «», inject text at :10 («Author → MERGE the kickoff to staging → only THEN hand out»).
• Companion principle test (CI): packages/core/principles/44-kickoff-authoring-traps.test.ts — arm A (symlink mode 120000 via git ls-files -s) + arm B (on-disk population) (:45-46).
• Earlier-channel gates: .husky/pre-commit CANON-symlink section (error) + check-kickoff-traps.sh arm 3 (:12, :45-46); SSOT for names is classifyKickoffName in packages/core/principles/kickoff-population.ts (:79).
• Not backend-rendered — no FF diagnostic applies (FF7001/FF7002, packages/core/diagnostics/registry.ts:321/:329); honest status: §1 merge-timing is prose + inject by design (:36 — «there is no pre-push/CI gate for «is this kickoff on staging yet?», because at author time the answer is legitimately «not yet»»).