Wish

/wish is one task delivered, as soon as possible. Give it a decided, bounded objective and it comes back with a merge-ready PR against the base branch, or with a refusal that names the route the request needs instead. Everything else in the lifecycle — brainstorm, plan review, work over many groups, fix, verify — is auxiliary to that.

Prefer the saved workflow: .claude/workflows/wish.js in the genie repository's workflow catalog (see .claude/workflows/README.md there) is authoritative for saved execution. On a runtime that runs saved workflows, run the script at <repository root>/.claude/workflows/wish.js when that file exists, otherwise ~/.claude/workflows/wish.js (delivered by genie install / genie update); give the runtime that explicit script path, never a bare name — both scopes now carry these names and the order a name resolves in is undocumented — and relay the returned result unchanged. If the runtime cannot run saved workflows or neither script exists, use the native fallback below; it needs neither script nor its runner.

Invoke

Pass {objective, issue?, context?, slug?, base?, repairBudget?, model?, gateModel?, publishModel?, check?, install?, validation?, timestamp} through args, or record them as coordinator input in native mode. objective is the one task, frozen: no stage re-asks, narrows or widens it. issue is a number or URL the scout reads and the PR links. context is frozen caller context (a file set, a decision reference) framed as data. slug names the branch wish/<slug> and the worktree .claude/worktrees/wish-<slug>; pass the brainstorm slug when one exists so the design preflight and the branch agree. base defaults to dev; anything that is not a plain branch name, and any spelling that resolves to the default branch (main, master, HEAD, origin/main, refs/heads/master), is rejected before any agent runs. repairBudget defaults to 2 and is capped at 3; pass the value of genie config get budgets.maxEscalationsPerGroup when the repository sets one. Saved execution uses the script's configured worker and reasoner tiers. model pins one model for every stage. gateModel and publishModel pin the two mechanical stages — the gate runs the repository check and reads an exit code, the publisher runs an allowlisted push/PR sequence — and win over model; the order is gateModel/publishModel, then model, then the tier. Native execution uses available role-appropriate models with that override precedence. The scout discovers the repository's check and install commands from its own root files (package.json scripts and the lockfile, or a Makefile, justfile or Taskfile target) and quotes what each one runs. The script, never the judge, writes them into the contract at the judge step, and freezes one empty when it is not exactly one of the forms the discovery rules produce, when the scout quoted nothing for it, or when its text or quoted body spells an obvious push, remote change or publish. In native mode the parent coordinator takes this discovery-validation and freezing responsibility, freezing the objective, scope and acceptance criteria before edits. check and install override them and face only that last test, on the command as written. It is a check against obvious push and publish commands, not a guarantee: it reads only the command and the quoted bodies one level deep, so a variable, an encoded string, a script the body calls, or a quote that does not match the file passes it. validation is the one command the gate runs verbatim when the repository has no hook system or no check command: without it the script freezes the judge's or the scout's answer to ONE focused test invocation over the declared file set — a runner from the closed list, no second command, and after the runner only files the declaration carries (a bare name with no / included) plus a tiny per-runner allowlist (--bail for bun test, -- for the script runners, run for npx vitest). Every other flag is refused, since a flag can load, configure or run code, and so are VAR=value, shell syntax, a .. segment, an absolute path and @file; a task runner (make test, just test, task test, rake test, tox, nox, mvn test, gradle test and the wrappers) takes no argument at all. That bounds what the command can name, not what the declared test files do when they run. validation overrides that answer, exempt from the shape and still checked as written. A refused answer stops the run at admission only when no check command froze (none discovered, or the one proposed was refused), because then the gate's only possible command is the validation command whatever the hook system; with a frozen check command, whether a refused validation command matters depends on a hook system only the gate classifies, so that stop stays at the gate. /wish runs the repository's own install and check commands, and an install such as npm ci runs every dependency's install scripts, so point it only at repositories you trust. timestamp is the caller's clock; the workflow has none.

Admission

The scout is read-only and estimates the work: files, insertions, independent units. The script (the parent coordinator in native mode) applies the size band from the wish-duration study — maximum 25 files and 2,000 insertions, the hard maxima; 3 units, advisory; ideal 10, 800, 2 — and a blind judge that never sees the repository decides the route: proceed, report (the cause is unknown), brainstorm (a product decision is open, or a linked design fails its preflight), or plan (too big, or the change touches a trust-boundary path: workflows, hooks, settings, release scripts, permission surfaces). Any route but proceed returns refused with nothing created. One more admission stop follows the route: when no check command froze and the validation command was refused, the run ends blocked at admission, before a worktree, a branch or an executor exists, because in that contract every hook mode ends at the same empty command. The reason names the refusal and the escape, a caller validation.

Relay

A run returns {ok, state, route?, contract, estimate, diff, head, branch, worktree, pr?, checks, review, gate, gateCommand, repairs, injectionAttempts, notConvened, scoutMikro, report}; gateCommand is the last gate's {command, mode} chosen from the contract (check, no-hook-system or no-check-command), null when no gate answered. Relay the saved workflow's report unchanged; in native mode the parent renders these same fields and state meanings from observed evidence. Run genie wish report <runId> --append and relay its output only when execution supplies a real report-compatible workflow run record and its runId. Native execution without that record explicitly reports accounting unavailable: observed evidence only, no fabricated run ID, token/time metrics or ledger append. state is one of:

• merge-ready — checks pass, the remote head equals the local head, the PR's base, head and file set equal the frozen contract, and the verdict is SHIP.
• pr-open — the PR exists but the checks had not concluded; re-read them with gh pr checks <n> before acting.
• refused — admission chose a route; nothing was created.
• blocked — at admission (a refused validation command with no check command frozen: no worktree, branch or executor exists), before publish (a worktree, dead hooks in a repository that has a hook system, no validation command in one that has none, a hook system with neither a check nor a validation command to run, a denylist hit or a BLOCKED review verdict), at publish (the host carries no gh, so nothing was pushed and no credential was sought), or at read-back after publish (a failing check or a structural mismatch, with the PR preserved and named); the reason names which. When a PR exists, inspect it before any rerun.
• missed — the repair budget ran out, or a stage threw or returned nothing; the branch, commit, worktree and any PR are preserved and named.

notConvened holds only agents that returned nothing; they were never counted as a pass. Merge, SHIPPED, dev→main promotion and worktree removal stay with the operator: after merge, git worktree remove <path> && git branch -d wish/<slug> (non-forcing, so an unmerged branch is refused). Retry after missed or blocked is a rerun with the same objective and slug; it adopts the named worktree when it is clean and nothing has diverged from the remote, and is otherwise blocked with a diagnostic that shows the unpushed work. The workflow never deletes a worktree or a branch.

Contracts it inherits

These clauses are the lifecycle's, restated nowhere else; the parity test pins them to the skills they come from, as it does the two rules the plan entry borrows from work and quotes there.

• From review: "The reviewer is different from the author and remains read-only."
• From fix: the repair budget is "default 2 when the key is unset".
• From work: "a worker notification is not delivery evidence by itself."

Plan a multi-group wish

The direct entry for work that is bigger than one task. Do not run admission; write the plan. An existing wish is resumed by editing it in place — the scaffold below refuses a destination that exists, and task rows and Run/Task/Dispatch identifiers already in use are reconciled before anything new is created. Use brainstorm when unresolved decisions prevent testable criteria. Write .genie/wishes/<slug>/WISH.md from the bundled template. Documents hold the plan and dependency DAG; the selected runtime holds execution state, and work executes the approved plan.

Before creating or changing a wish that links a brainstorm, run the design preflight in references/design-preflight.md and take its outcome as final. Missing evidence, a non-SHIP verdict, or a content-digest mismatch cannot be waived. Never repair the failure with a locally recomputed digest.

Scaffold and fill

set -eu
WISH_SKILL_DIR='<absolute directory containing this SKILL.md>'
WISH_SLUG='<slug>'
case "$WISH_SLUG" in
  ''|*[!a-z0-9-]*|-*|*-) printf 'invalid wish slug: %s\n' "$WISH_SLUG" >&2; exit 2 ;;
esac
WISH_DEST=".genie/wishes/$WISH_SLUG/WISH.md"
test -f "$WISH_SKILL_DIR/templates/wish-template.md"
test ! -e "$WISH_DEST"
mkdir -p "$(dirname "$WISH_DEST")"
cp "$WISH_SKILL_DIR/templates/wish-template.md" "$WISH_DEST"

Fill {{slug}}, {{date}}, and every TODO. Preserve the template’s machine-consumed structure exactly: the # Wish: title, the metadata table rows, and the sections ## Summary, ## Scope with ### IN and ### OUT, ## Decisions, ## Simplicity Case, ## Dependencies, ## Success Criteria, ## Execution Strategy, ## Execution Groups holding at least one ### Group <n>: heading, ## QA Criteria, ## Assumptions / Risks, ## Review Results, and ## Files to Create/Modify. Inside every group keep the **Goal:**, **Deliverables:**, **Interfaces:**, **Acceptance Criteria:**, **Validation:**, and **depends-on:** blocks, and keep the Execution Strategy columns including Complexity and Model. Use portable roles/reasoning effort in the plan; runtime configuration selects actual models.

Pass the simplicity gate: state the smallest complete design, justify added machinery with present requirements or measurements, and keep deferred mechanisms out of execution. Give each group a goal, owned files, deliverables, testable criteria, dependencies, and a non-zero validation command; the repository’s required gate is sufficient rationale and a group’s own checks never replace it: preserve repository-required aggregate gates. Parallel writers need disjoint file ownership or dedicated worktrees; otherwise sequence them. Size the plan by the study: a group that would exceed the admission band is a sibling wish, not a bigger group. Declare wish-level **depends-on:** and **blocks:** under ## Dependencies (comma-separated slugs or none), plus per-group **depends-on:**. Fill each group's **Interfaces:** block with exact signatures, and copy the plan-wide requirements into **Global constraints:** verbatim. A wide refactor sequences expand, migrate, contract, with green promised only in a final integrate-and-verify group.

Review and handoff

1. Run the wish linter over this repository’s own .genie/wishes, in any repository:

genie wish lint

It reports structure only — template sections, the Status and Date metadata, the Execution Strategy routing columns, brainstorm links that resolve — writes nothing, and exits 0 clean or 1 with findings; --dir <repo> names another checkout. In the Genie repository the same linter is also a required stage of the repository gate, reached through its alias grep -q '"wishes:lint"' package.json 2>/dev/null && bun run wishes:lint. A linter the project does not provide is reported as a finding, and a linter that runs and fails blocks handoff. 2. Obtain independent review of the completed plan. The caller appends its evidence under ## Review Results and persists APPROVED, FIX-FIRST, or BLOCKED. work requires APPROVED on disk. 3. Create missing task rows per group (genie task create --title "<group title>" --wish <slug> --group <group-name>) and inspect for duplicates before retrying; an unavailable CLI is reported, never bypassed. 4. After APPROVED, run genie context --wish <slug> to record the wave base SHA.

Without a workflow surface

Load references/native-fallback.md relative to this loaded SKILL.md before dispatch. It is the standalone execution guide for Admit, Work, Gate, Review, Repair, Publish, Read-back and Render, sufficient when the saved scripts and their runner are absent. The parent is the coordinator: dispatch native subagents with explicit briefs and prior outputs, enforce dependencies and transitions, and render the result; do not build another saved workflow. Follow the guide's discovery, frozen-contract, gate, independent-review, bounded-repair and publication safeguards, including SHIP plus a green gate before publish. Its enforcement is procedural, not script-level automation; accounting follows Relay above.