Gabe Execute

Gabe Execute — phase implementation runner

Usage: /gabe-execute [task|all|<phase-number>] [--auto-commit] [--dry-run]

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

Executes phase tasks from .kdbp/PLAN.md. Complements /gabe-plan (write plan) and /gabe-commit (quality gate) — this is the implementation step: reads the plan, writes code, checkpoints at commit boundaries (user-gated by default, --auto-commit batches), and advances Exec state. Default scope is the single Current Phase; task runs one task only, all runs remaining phases autonomously, <N> jumps to phase N.

Procedure

1. Treat any text after the invocation as $ARGUMENTS.
2. Read references/execute-spec.md (in this skill directory) IN FULL before executing — it is the binding spec: task decomposition, tier-cap enforcement, escalation gate, commit invocation, deviation handling, and phase-complete invariants. If it is missing, E6 applies — STOP.
3. Summary of the spec's main flow:
   • Step 0 — parse args (empty/task/all/<N>/--auto-commit/--dry-run); preconditions: .kdbp/ exists, PLAN.md is active with an Exec column; project-type preflight redirects mockup plans (and mockup-typed phases of hybrid plans) to /gabe-mockup instead of proceeding.
   • Step 1 — load execution context: Current Phase row (name, tier, complexity, Exec state, per-dim tier overrides from Phase Details YAML), BEHAVIOR.md maturity/execute mode, PENDING.md open items for this phase's files, tier-cap heuristics, and a deterministic classification of whether runtime journey evidence (and staging proof) is required for this phase's types.
   • Step 2 — decompose the phase into tasks via deterministic heuristics first (comma/semicolon-separated actions, distinct Scope files, multiple References specs), falling back to Haiku only when heuristics yield <2 or >10 tasks; prune tasks above the effective tier unless justified; append a mandatory runtime-journey-evidence task when required.
   • Step 3 — tick Exec ⬜ → 🔄.
   • Step 4 — execute tasks one at a time with per-task verification.
   • Step 4.1 — mid-phase tier escalation gate: fires on user request, a task genuinely needing a higher-tier pattern, or a drift signal; requires a mandatory one-sentence reason, updates PLAN.md's Tier cell, appends a Tier escalation entry under the phase's DECISIONS.md D-entry, reinstates pruned tasks, and logs a thin-index row to LEDGER.md. De-escalation is not supported mid-phase.
   • Step 4.5 — commit checkpoint: MUST invoke /gabe-commit inline (raw git commit prohibited) so CHECK 6/7/8, the LEDGER thin-index row (findings/deferred/size-budget), PENDING.md updates, the teach suggestion, and the Commit-column auto-tick all fire; CRITICAL findings block advancing to the next task; confirms the Commit cell actually ticked before continuing.
   • Step 5 — commit message enrichment (gabe-lens brief + before/after).
   • Step 6 — deviation handling: structural deviations halt with an update-plan/split-task/skip-task/abort prompt; minor deviations are recorded in the checkpoint commit body and counted in the EXEC LEDGER row's deviations cell, and execution continues.
   • Step 7 — phase complete: verifies the runtime-journey-evidence invariant (artifact paths must exist on disk, checked via ls, not claimed in prose) and the Commit-column invariant before ticking Exec 🔄 → ✅ via the shared auto-tick helper.
   • Step 8 — interrupts/resume: an aborted phase stays 🔄; the next invocation reads completed/remaining tasks from the PLAN.md tasks block, never from session memory.
   • Model routing: Haiku for decomposition fallback and deviation classification, Sonnet for implementation and conceptual commit messages. Non-goals: does not replace /gabe-commit or /gabe-review (invokes/surfaces via them), does not auto-push, does not write architectural docs.

Output contract (summary)

Report per-task progress (files changed, verification result, commit hash) and end each phase with the Phases-table state line (EXEC: … REVIEW: … COMMIT: … PUSH: …) plus a Next: pointer (typically /gabe-review or /gabe-next). Emit the output-only **Gabe-Lens block** and **Gabe-Lens brief — Platform progress** — never written to PLAN.md/REVIEW.md/LEDGER.md/PENDING.md/commits/docs unless another command already owns that write. All state writes (Exec ticks, DECISIONS.md escalations, PLAN.json proof mirrors, LEDGER.md thin-index rows) happen in the same turn as the action they record (E5); a skipped write prints an enumerated skip code, never silence. The full output contract in the spec is binding.

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.