Execute Common

Execute-family controller recipe

One home for controller steps and runtime state shared across build-in-waves, build-by-story, and build-inline. Task dispatch/review lives in task-lifecycle.md beside this file. Load when that skill's Setup or After-last says to. Registered so npx skills add copies it with the execute family.

Session preflight

Three questions, before any dispatch, any isolate-workspace / git worktree add, and any first production edit:

1. Issue tracker sync. WHEN docs/agents/issue-tracker.md names a configured tracker, read tracker-sync.md beside this file and follow it exactly; otherwise → empty ticket set, continue (unconfigured is normal, not a failure).

2. Catalog occupancy. Read Status: from requirements.md and the matching docs/specs/INDEX.md row. INDEX missing → say so once, suggest /configure-repo; still write requirements.md when that file exists. land-branch / realign-spec own Implemented; cut-release owns Shipped. Runs on every start, including compaction resume.

   Status	Action
   Approved	On this checkout, write In-progress on both files, then continue.
   In-progress	Already occupied. Do not re-stamp. Continue.
   Draft	STOP — spec is not approved.
   Implemented / Shipped	Leave it. Not a kickoff stamp.

   Before isolation, commit this checkout's dirty spec files for the feature (requirements.md, design.md, tasks.md, the INDEX row) — the occupancy write is part of that commit when Status was Approved. A worktree is created from HEAD, so uncommitted spec docs stay behind, invisible to the new tree and other sessions. isolate-workspace still must not commit .gitignore — occupancy is this step, not that skill.

3. Workspace / branch. If no isolated workspace exists yet: isolate in a worktree, or implement on the current branch? Do not create a worktree unasked. Isolation → REQUIRED SUB-SKILL: use isolate-workspace. Current branch is main/master → separate explicit consent before implementing; "no worktree" is not consent to touch main/master.

Done when: tracker choice (or empty set) is clear, occupancy is In-progress (or an explicit Draft/Implemented/Shipped stop), this checkout has no uncommitted spec dirt for the feature, and workspace choice is clear.

Runtime binding and lease preflight

Before the first dispatch (inline route: before the first production edit), read runtime-binding.md beside this file and follow it exactly. It defines the .skills/<CODE>/execution-session.json schema (schema_version, harness, provider, model, resume/fork/worktree support, cache_control, token_telemetry, pricing_policy, effective_concurrency, rotations), the four lease-rotation triggers, the pricing-policy fallback, and the concurrency-degradation rule.

Done when: the runtime snapshot exists before dispatch, every unavailable capability is explicit, and the first effective concurrency/lease decision is recorded.

Ledger check

Read ledger-check.md beside this file and follow it exactly: it keeps .skills/ local-only, resumes from .skills/<CODE>/progress.md, and defines the Verified: completion-claim slot backed by prove-claim.

Done when: next task / unit is known.

Decision trail — observable conditional

WHEN user asked for a decision trail / show-work, OR unattended overnight / multi-day, OR effective_concurrency > 1 with multiple open units → read decision-trail.md beside this file and follow it exactly; ELSE write skip: no decision-trail predicate on Close notes.

Todos — GATE

Before any dispatch or Task 1, on a visible list — the harness's todo / task-list tool when it exposes one, otherwise a checklist written into your reply and restated at each task boundary: one todo per task and one terminal todo Close branch (the close sequence below — created now, not later). Add a Polish Diff todo only when a polish predicate is already known to hold (user asked up front). Otherwise create the Polish Diff todo at close-sequence step 3 if a predicate then holds.

Done when: the list mirrors the plan and includes the Close branch todo.

Close sequence

After the last task (waves / inline) or last unlocked unit (story):

1. Whole-branch review. REQUIRED SUB-SKILL: use inspect-change with base = git merge-base main HEAD — never a mid-branch sha. Feed ledger Minors. Top model tier.
2. One fixer for the complete findings list → re-review. Never one fixer per finding. Inline route: you are the fixer under test-first.
3. Polish Diff — observable conditional. Evaluate the polish predicate below. IF any clause is true → REQUIRED SUB-SKILL: use polish-diff on the whole-branch diff before acceptance; create the Polish Diff todo now if it does not exist; mark it done only after the skill has run. IF no clause is true → skip; write skip: no polish predicate on the Close branch notes. EOD, demo pressure, "inspect was clean", and "small enough to feel optional" are not predicates.
4. Acceptance. REQUIRED SUB-SKILL: use validate-feature. Breaks → root-cause, then promote passing checks to committed tests that describe the domain behavior (docs-only spine — no requirement-ID tags required in test files).
5. Sample — observable conditional. Evaluate the sample predicate below. Any clause true → write sample: required on the Close notes; none true → write skip: no sample predicate. Do not start /select-sample. A silent skip is still a red flag, and "inspect was clean" / "always name so we cannot forget" are not predicates.
6. Product walk — observable conditional. Evaluate the walk predicate below. No clause true → do not open the product-walk trio. Any clause true → REQUIRED SUB-SKILL: use write-flow-guide (it owns vet), then execute what it wrote — REQUIRED SUB-SKILL: use run-flow-guide where the runtime can drive the app, else record the human ticks. Naming the run instead of doing it leaves step 7 unable to write green.
7. Close receipt. Load close-receipt.md and follow its producer recipe. Issue it only after the final mutation and only from evidence bound to that HEAD. Rerun a producer only when its evidence is missing or stale.
8. Finish. REQUIRED SUB-SKILL: use land-branch.

Mark the Close branch todo done only after steps 1–8 have each run or been skipped under their predicate.

Close-sequence predicates

Polish — true when any of: the user asked for polish, cleanup, or tidy; inspect-change Standards leftovers include an Important-or-higher finding that is behavior-preserving (reuse, dead code, needless complexity, wasted I/O); git diff --name-only $(git merge-base main HEAD) HEAD lists more than 15 files; the branch adds a new public API or exported surface.

Sample — true when any of: the user asked for a sample or attention allocation; branch diff paths hit the B1 defaults in skills/review/select-sample/references/signals.md extended by project Risk globs; the branch diff lists more than 15 files. "Always name so we cannot forget" is not an ask.

Product walk — true when any of: the user asked for a product walk, dogfood, or walkthrough; validate-feature reports neither-API-nor-UI; an approved requirement uses visual / feel / eyeball language the automated surfaces cannot judge; inspect-ui's report (via inspect-change's UI lane) lists any needs-human-eyes item; the branch adds a new user-facing screen or visual surface — not only changes within existing ones.

Red flags — never

• Skip the tracker-sync, occupancy, or workspace preflight
• Isolate or edit production files while catalog Status is still Approved
• Write In-progress only inside a worktree created this session
• Re-stamp an already In-progress row
• Isolate while this checkout still has uncommitted spec/requirements dirt
• Invent a tracker or ticket set when config is absent or the user declined
• Start implementation on main/master without explicit consent
• Create a worktree without asking, or treat "current branch" as consent for main/master
• Dispatch the first task before the todo list exists (tasks and Close branch)
• Silent-skip polish (no written skip: no polish predicate)
• Silent-skip sample or decision-trail (no written skip / required line)
• Start /select-sample in this skill
• Treat EOD, demo, or "inspect was clean" as a polish predicate
• Move to land-branch with the Close branch todo still open
• Write a partial receipt or bind it to evidence from before the last mutation
• Promote checks as "ID-tagged tests" — IDs stay in docs