Gabe Push

Gabe Push — env-aware shipping workflow

Usage: /gabe-push [env-name] [--epic] [--reconfigure]

Gabe execution contract (E1–E7)

This skill runs under the suite execution contract — E1 EVIDENCE · E2 RUN-BEFORE-✅ · E3 NO SILENT DOWNGRADE · E4 REUSE FIRST · E5 STATE SYNC · E6 MISSING ANCHOR = STOP · E7 REPORT WHERE — floors, not ceilings; a skill's own gate may be stricter, never looser. Full text: ../gabe-docs/references/execution-contract.md (if that file is missing, E6 applies — STOP).

What this does

Env-aware shipping. One command pushes local work to the configured target env, or promotes what's already on a pre-prod env (e.g., staging) up to the next env (e.g., main/production). Config lives in .kdbp/PUSH.md; first run interviews for envs (production-only, staging-then-prod, or custom N-env), subsequent runs honor the config until /gabe-push --reconfigure. Every run detects remote branch drift and offers cleanup of the source branch after a successful push.

Procedure

1. Treat any text after the invocation as $ARGUMENTS.
2. Read references/push-spec.md (in this skill directory) IN FULL before executing — it is the binding spec: PUSH.md setup, env resolution, drift detection, push/PR/CI logic, DEPLOYMENTS.md capture, bookkeeping auto-commit, and branch cleanup. If it is missing, E6 applies — STOP.
3. Summary of the spec's main flow:
   • Step 1 — validate gh CLI + auth + git remote; read .kdbp/PUSH.md if present.
   • Non-interactive defaults — when no human answer is available, safety-critical prompts (uncommitted changes, direct-push-to-target, CI failure) default to abort/stop, never proceed.
   • Step 2 — first-run setup: detect remote/default-branch/CI provider/PR template, ask the deploy pattern (production-only / staging-then-prod / custom), scaffold envs, ask branch_cleanup policy per env, write .kdbp/PUSH.md. --reconfigure clears and re-runs this step.
   • Step 2.5 — resolve which env this invocation targets from $ARGUMENTS (default env, named env, or interview for an unknown env).
   • Step 2.7 — remote branch drift detection every run: diff current remote branches against known set, prompt per unrecognized branch (ignore-once/ignore-always/register-as-env/delete-remote/abort), persist non-transient decisions.
   • Step 3–4 — determine push source (promote from promote_from vs push local HEAD), pre-flight checks, then push.
   • Step 5 — create or update the PR (skipped for direct pushes to the target branch or remote-to-remote promotion).
   • Step 6 — CI watch, non-blocking, up to 75s (5 polls × 15s); on failure offers details/logs/auto-fix/assess/ignore; ⏳ is never treated as ✅.
   • Step 6.7 — deploy verify: live-target smoke probe is the closing evidence, not CI-green alone.
   • Step 7 — final summary; promotion to a further env only happens on a separate invocation (never recurses across envs).
   • Step 7.5 / 7.5b — append a row to .kdbp/DEPLOYMENTS.md (deterministic aggregation, zero LLM) and run the operational-decision classifier (Haiku, trigger-gated on CI-config/infra/deploy-config changes, rollback commits, trunk-first pushes, skipped promotions) with interactive triage.
   • Step 8 — record to .kdbp/LEDGER.md.
   • Step 8.5 — auto-commit post-push bookkeeping (PUSH.md/DEPLOYMENTS.md/LEDGER.md/DECISIONS.md/PENDING.md writes) as a local, unpushed commit through the normal hook chain — never git add -A, never a hook bypass.
   • Step 9 — retired (was a legacy /gabe-teach topics suggestion; the skill is archived and KNOWLEDGE.md left the default inventory — always a no-op now).
   • Step 10 — auto-tick the Push column in PLAN.md via the shared auto-tick helper (/gabe-plan's "Shared: auto-tick phase column"), only when push succeeded, CI is green, the env is the configured final environment, and promotion reached the final link; prints the decision record before ticking or skipping.
   • Step 10.5 — branch cleanup prompt (delete/keep/always/never) after a successful push, skipped for promotion pushes and direct-to-target pushes.

CI babysitting (runbook)

For long CI runs after push, a mechanical watch loop may babysit the pipeline instead of holding the session open — ONLY where the project's verification gates are in place; unattended auto-mode without verification is not allowed. Use /loop (e.g. /loop 4m check CI on <branch> and report) for watch-and-report only; auto-fix loops are appropriate only where the phase has runtime-journey proof in place (PLAN.json proof).

Output contract (summary)

End with the GABE PUSH COMPLETE summary block (env, source, target, PR url or —, CI status) and a Bookkeeping: line reporting whether the bookkeeping commit happened. All state writes (PUSH.md, DEPLOYMENTS.md row, LEDGER.md entry, PLAN.md Push tick, PENDING.md/DECISIONS.md from the operational classifier) land in the same turn as the action they record (E5); a skipped Push tick prints the full decision record, never a silent skip. Safety-row prompts never silently default to proceed. The full output contract in the spec is binding.

Emit the shared beat brief (**Gabe-Lens brief** — ENTITY / FEATURE / DID, stated once in ../gabe-docs/references/execution-contract.md §"The beat brief") just before the E8 tail; output-only, never persisted.

Closing — the beat tail (E8)

End every run with the three-part beat tail, specified ONCE in ../gabe-docs/references/execution-contract.md §"The beat tail (E8)": NOW:/NEXT: rendered from node ${ECC_ROOT:-$HOME/.claude}/skills/gabe-next/scripts/next.mjs --json (or the honest NEXT: blocked — <reason> override when this beat knows the router's answer is stale) · the conditional CENTER: pointer · the PULSE line last, verbatim, silent when silent.