| name | feature |
| description | Start a new feature in an isolated git worktree with proportional verification |
| when_to_use | when the user says "start a feature", "new feature", invokes "/agent-dashboard:feature", or describes work that needs an isolated branch + worktree + planned implementation loop. NOT for hotfixes, single-file edits, pure exploration, or non-code changes (use /agent-dashboard:chore, /agent-dashboard:fix, /agent-dashboard:investigate instead). |
| version | 1.0.0 |
| disable-model-invocation | true |
| effort | max |
<codex_skill_must>
- Phase 2 is gated on
permission_mode='plan'; if not plan, stop and ask the user to run /plan.
- Run
mkdir -p and git worktree add -b feat/<name> ... main as separate exec_command calls.
- Submit plans only inside
<proposed_plan>...</proposed_plan>.
- After
<proposed_plan>, stop until plan-review approval; ordinary chat approval is not implementation approval.
- After approval, write
.feature-plan-path before implementation.
- Allowed tools:
exec_command, request_user_input, spawn_agent, wait_agent, update_plan, apply_patch; every spawn_agent needs wait_agent.
</codex_skill_must>
Start a new feature in an isolated git worktree.
Feature description: $ARGUMENTS
Instructions
Follow these phases in order. Each phase has a gate — do not proceed until the gate is satisfied.
Before every action, identify the current phase and check its gate. If a gate is not satisfied, stop instead of falling back. If you violate phase order, halt and report the violated gate.
If the feature touches browser UI, Playwright, dev-server ports, screenshots, or interactive Browser/Chrome inspection, apply ../_shared/ui-automation.md at planning, environment setup, verification, delegation, and cleanup points.
For Verification profiles, apply ../_shared/verification-profiles.md. Active AGENTS.md/core rules may add doctrine, but this shared glossary is the standalone agent-dashboard fallback.
Phase 1: Setup
Follow ../_shared/worktree-setup.md with branch prefix feat.
Gate: Working directory is the new worktree on the correct branch, based on latest main. If .env* files existed in the source repo, they are all present in the worktree.
Phase 2: Plan
Start two tracks in parallel:
Plan-first guarantee. Phase 2 is planning only. No dependency installs, no
sentinel writes, no environment setup runs until the plan is approved. The
only writes allowed in this phase are read-only research (Codex explorer)
and the <proposed_plan> submission itself. Environment setup kicks off in
the post-approval actions below.
Background — Environment setup: Launch a background agent (run_in_background: true) to set up the dev environment per ../_shared/env-setup.md.
Phase order: Plan Mode first, then research, then interview, then submit. Plan Mode is the entry gate to Phase 2 — nothing else in this phase happens outside it. Each step has a HARD-GATE you cannot rationalize past.
-
Enter Codex Plan Mode via /plan before continuing. Dashboard-created Codex feature sessions should auto-enter Plan Mode before the skill prompt is submitted. If permission_mode is not plan, stop immediately and ask the user to run /plan; do not research, interview, draft, call spawn_agent, call update_plan, edit files, write tests, run setup, or create sentinels.
No spawn_agent, request_user_input, or plan drafting until permission_mode='plan'.
-
Research with spawn_agent explorer. Use a Codex explorer subagent for non-trivial codebase or library lookups. Always pair spawn_agent with a wait_agent to retrieve the explorer's findings — never spawn-and-forget. Do not delegate planning — composing the plan is your job. Synthesize what you found inline as your own assistant text so it lands in the visible plan artifact (not in a tool result the user can't approve).
explorer for research is fine; never dispatch another role to compose the plan. Every spawn_agent must be followed by wait_agent.
Foreground — Planning:
- Interview the user via
request_user_input. Identify every gating decision the implementation depends on — URLs, IDs, scope boundaries, copy text, what to delete vs keep, version pins, credentials. Ask them as a single request_user_input call with multi-choice options, not as freeform numbered text. If request_user_input is unavailable during Phase 2, stop and ask the user to run /plan.
Phase order: research first, interview second, plan mode third, submit fourth. Plan mode is the last gate before approval, not a pre-research speed-bump. Each step has a HARD-GATE you cannot rationalize past.
-
Research with Explore. Use the built-in Explore subagent for any non-trivial codebase question or library lookup. Do not call Agent with subagent_type=Plan — composing the plan is your job, not a delegated subagent's. Synthesize what you found inline as your own assistant text.
Research subagent (`Explore`): allowed and encouraged.
Planning subagent (`Plan`): forbidden in this skill. Compose the plan yourself.
Do not wait for environment setup to finish.
-
Interview the user via AskUserQuestion. Identify every gating decision the implementation depends on — URLs, IDs, scope boundaries, copy text, what to delete vs keep, version pins, credentials. Ask them as a single AskUserQuestion call with multi-choice options, not as freeform numbered text in your assistant message.
Load via ToolSearch if AskUserQuestion isn't in scope: ToolSearch("select:AskUserQuestion").
Schema: 1–4 questions per call, each with a header (≤12 chars), 2–4 mutually exclusive options, multiSelect: false unless the choices genuinely combine. Recommended option goes first with (Recommended) suffix. The user always gets an "Other" escape hatch automatically.
Schema: 1–3 questions per call, each with a header (≤12 chars), 2–3 mutually exclusive options. Recommended option goes first with (Recommended) suffix. The client adds an "Other" escape hatch automatically.
Worked example:
<!-- claude-only -->
AskUserQuestion({
<!-- /claude-only -->
<!-- codex-only -->
request_user_input({
<!-- /codex-only -->
questions: [{
<!-- codex-only -->
id: "focus_path",
<!-- /codex-only -->
question: "Where should focus.json live?",
header: "Focus path",
<!-- claude-only -->
multiSelect: false,
<!-- /claude-only -->
options: [
{label: "~/.agent-dashboard/focus.json (Recommended)", description: "Co-located with agents/ state dir already watched."},
{label: "$XDG_RUNTIME_DIR/agent-dashboard/focus.json", description: "Tmpfs-backed, cleared on reboot."},
],
}]
})
The plan you submit in step 4 must be implementable as written. No "Decisions needed", "Phase 0", "TBD", "?", or "to be confirmed" sections in the body. If a user answer changes scope, return to this step and re-interview.
The plan you submit in step 4 must be implementable as written. No "Decisions needed", "Phase 0", "TBD", "?", or "to be confirmed" sections. If a user answer changes scope, re-interview before submitting.
If you find yourself typing "1." "2." "3." questions in assistant text — STOP. Use request_user_input.
Freeform numbered questions in assistant text are a violation. If you find yourself typing "1." "2." "3." to ask the user something — STOP, call `AskUserQuestion` instead.
The plan is not ready for review until every decision it gates is answered.
4. **Draft inline, then submit via ``. Wait for user approval.** Plan Mode is already active (step 1); draft the plan as your own assistant text and then wrap the full plan markdown in `...`. This renders the plan through Codex's native plan-review flow for accept/reject. After submitting ``, stop until the plan-review approval arrives; ordinary chat approval is not implementation approval.
- Enter plan mode via
EnterPlanMode, then draft the plan inline. Now that research is done and decisions are resolved, call EnterPlanMode (load via ToolSearch if not in scope). This flips the parent's permission_mode='plan' and restricts you to read-only tools while you write the plan as your own assistant text.
Caveat: update_plan is a progress checklist tool, not a planning-mode substitute. Use it after approval to track implementation progress; do not use it to bypass Plan Mode.
Caveat: on approval (step 4), CC drops to its default permission_mode, not back to bypassPermissions. Subsequent edits in Phase 3 will re-prompt unless the user re-enables bypass. Accepted trade-off — visible planning is worth the one-time mode reset.
No drafting the plan in assistant text until `EnterPlanMode` has been called and `permission_mode='plan'` is active.
Phase format for multi-phase plans. If the plan has 3+ distinct work units, structure it with a ## Phases checklist and matching ### Phase X: headings. Step 4 reads this format to offer the dispatch probe; /agent-dashboard:implement parses it to drive the dispatch loop. Plans without it can't be dispatched.
Phase format for multi-phase plans. If the plan has 3+ distinct work units, structure it with a ## Phases checklist and matching ### Phase X: headings. The post-approval probe reads this format to offer dispatch; /agent-dashboard:implement parses it to drive the dispatch loop. Plans without it can't be dispatched.
## Phases
- [ ] **Phase A: <short name>** — files: <globs>, deps: -
- [ ] **Phase B: <short name>** — files: <globs>, deps: A
- [ ] **Phase C: <short name>** — files: <globs>, deps: B
### Phase A: <short name>
<10–50 lines: what files, Verification profile (Surgical/Targeted/Full) + proof command, what invariants, what to leave alone.>
### Phase B: <short name>
<...>
Rules:
- The
## Phases block is the dispatch index. Phase names MUST match between checklist and ### Phase X: headings (case-sensitive).
- Each phase body MUST name a Verification profile and proof command so
/agent-dashboard:implement does not default to whole-repo tests for isolated work.
deps: defaults to "depends on previous phase". Use - for "no dependencies".
- [ ] = pending. - [x] = done. /agent-dashboard:implement flips these as it dispatches.
- Fewer than 3 work units? Skip this format. Inline paragraphs are fine; the probe won't fire below the threshold.
-
Submit via ExitPlanMode. Wait for user approval. Pass the full plan markdown to the plan file (per CC's plan-mode workflow) and call ExitPlanMode. This renders the plan in CC's native plan-review UI for accept/reject.
ExitPlanMode is the only acceptable submission. Pasting the plan as assistant text is a violation, even if you also call ExitPlanMode afterwards. The user reviews and approves through the plan-review UI — nowhere else.
<proposed_plan> is the only acceptable submission. Pasting the plan as ordinary assistant text is a violation, even if you also update a checklist afterwards. The user reviews and approves through the plan-review UI — nowhere else.
No exceptions:
- Don't start a "small" preparatory edit while waiting.
- Don't write the test file "to save time".
- Don't ask "should I proceed?" in assistant text — the plan-mode UI's accept action is the approval.
- Don't ask "should I proceed?" in assistant text — the plan-review UI's accept action is the approval.
Post-approval actions (immediately after the user accepts the plan, before Phase 3):
- Write the plan-path sentinel (always). CC's plan-mode system prompt told you where the approved plan markdown lives (typically
~/.claude/plans/<slug>.md). Record that path so /agent-dashboard:implement can find it:
- Write the plan-path sentinel (always). Save the approved plan markdown to a worktree-local file and record that absolute path so
/agent-dashboard:implement can find it:
```bash
echo "<absolute-plan-path>" > .feature-plan-path
```
- Count the phases. Read the plan; count
- [ ] / - [x] lines under ## Phases. If there's no ## Phases block or the count is < 3, skip step 3 below and start Phase 3 inline.
-
Kick off environment setup as a background exec_command (always). The setup task runs in parallel with the dispatch probe and any pre-Phase-3 work so its install time is amortised. It must follow ../_shared/env-setup.md.
-
Count the phases. Read the plan; count - [ ] / - [x] lines under ## Phases. If there's no ## Phases block or the count is < 3, skip step 4 below and start Phase 3 inline.
- Probe for dispatch handoff (only when phase count ≥ 3). Call
AskUserQuestion exactly once:
- Question:
"Plan has {N} phases. Continue inline here, or hand off to /agent-dashboard:implement for context isolation?"
- Probe for dispatch handoff (only when phase count ≥ 3). Call
request_user_input exactly once when available; otherwise ask one concise direct question and wait for the user's answer. Never choose the recommended option yourself.
- Question:
"Plan has {N} phases. Continue inline here, or hand off to /agent-dashboard:implement for context isolation?"
- Header: `"Dispatch"`
- Options (recommended first):
- `"Continue inline (Recommended for ≤4 phases)"` — Stay in this session; run each phase with its selected Verification profile and proof command.
- `"Hand off to /agent-dashboard:implement"` — Exit /agent-dashboard:feature. The user invokes `/agent-dashboard:implement` in a fresh session; each phase dispatches to its own subagent.
**If `Continue inline`:** start Phase 3. The `## Phases` structure becomes documentation — inline implementation ignores the index.
**If `Hand off to /agent-dashboard:implement`:** print the message below and exit cleanly. Do not start Phase 3.
```
Plan saved to <plan-path>.
Worktree ready at <worktree-path>.
To continue, run:
/agent-dashboard:implement
(Recommended: open a fresh terminal session for max context isolation.)
```
Gate: Plan approved with no open decisions. .feature-plan-path written. Either Phase 3 begins (inline) or the skill exited with the handoff message (dispatch).
Phase 3: Implement
Pre-gate: Check for .env-setup-done in the worktree root.
- If present: verify dependencies are installed (e.g.
node_modules/ exists, pip list succeeds, go env GOPATH works) and data symlinks resolve correctly.
- If
.env-setup-failed exists: surface the error and halt.
- If neither file exists: the background agent is still running — wait for it to finish before proceeding.
Effort note: When launched via the agent-dashboard's New Agent flow, this skill spawns with --effort high on the CLI, which Claude Code pins at the session level. The dynamic dispatcher in agent-state-fast.js bumps effort to max automatically while permission_mode='plan' (EnterPlanMode active) and drops back to high on exit — so planning runs at max effort without paying that cost during implementation. When invoked as a slash command inside an existing claude session, you can run /effort max before entering plan mode and /effort high (or lower) before implementation.
Effort note: When launched via the agent-dashboard's New Agent flow, this skill starts at implementation effort. Codex Plan Mode is the high-reasoning planning surface; once the approved plan is accepted, use update_plan for progress tracking and keep implementation effort proportional to the work.
Delegation gate: Invoke /codex:setup to check Codex CLI availability. If the output contains "ready": true, delegate only if the user explicitly requested Codex delegation OR the plan touches 10+ files / ~3,000+ lines of implementation. Below that threshold, the orchestration overhead (skill loading, prompt construction, subagent context, result parsing, review) costs more tokens than Claude implementing directly. If delegating, invoke /codex-delegate with the approved plan (Phase 2) as implementation context, then skip to the phase gate. Otherwise, proceed below.
Delegation gate: Use Codex spawn_agent only if the user explicitly requested subagents OR the plan touches 10+ files / ~3,000+ lines of implementation. Below that threshold, the orchestration overhead (skill loading, prompt construction, subagent context, result parsing, review) costs more tokens than implementing directly. If delegating, use /agent-dashboard:implement with the approved plan (Phase 2) as implementation context, then skip to the phase gate. Otherwise, proceed below.
Build the feature using the active AGENTS.md/core proportional verification doctrine when present, and ../_shared/verification-profiles.md as the standalone fallback. This skill selects the profile, records the proof command, and runs that command.
Loop:
- Profile first. State
Verification profile: <Surgical|Targeted|Full> and Proof command: <command or none> before editing. Escalate the profile if the diff grows.
- RED where it adds value. When the selected profile calls for behavior or regression coverage, write the smallest failing test first and show the failing output. Do not add implementation-only tests; when no new test is warranted, name the existing proof or why no executable proof applies.
- GREEN. Write the minimum implementation and rerun the proof command until it passes. No "while I'm here" additions or premature abstractions.
- REFACTOR. Clean up only structure. Rerun the proof command after meaningful edits; escalate if the refactor crosses package boundaries or changes shared behavior.
- Final implementation proof. Before committing from this skill, run the profile's proof command. The PR skill owns the final branch-wide cleanup, formatting, and full test gate.
For UI verification, prefer headless Playwright with worktree-local resources. Use interactive Browser/Chrome inspection only when the shared policy says it is warranted.
Gate: Environment ready. The selected verification profile passes. Implementation matches the approved plan, and unnecessary new tests were not added.
Phase 4: Review
Review all changes for correctness, security, and convention adherence. Apply all project rules and conventions that are in your context.
Gate: No critical or high-severity issues remain.
Phase 5: Deliver
- Commit the feature changes with a
feat: conventional commit message.
- Open the PR by invoking
/agent-dashboard:pr. That skill owns conditional cleanup/formatting, final test gating when available, push, and gh pr create. Do not call gh pr create directly — a pr-skill-gate hook will block it.
Gate: Clean commit history with conventional commit messages. PR opened via /agent-dashboard:pr.
Phase 6: Cleanup (on merge)
Triggered when the user indicates the feature has been merged upstream.
- Verify the branch is merged (warn if unmerged commits remain)
- Tear down environment resources: remove symlinks, stop dev servers or emulators, release any browser lease, remove worktree-local UI scratch state, delete
.env-setup-done / .env-setup-failed / .feature-plan-path sentinel files
- Remove worktree and delete branch
- Confirm cleanup is complete
Red Flags — STOP
If you catch yourself saying or thinking any of these, pause and re-read the relevant phase. (Hooks block the common paths for commits on main and direct gh pr create, so those get no self-check bullets here.)
- "I'll just sketch the implementation first" → Phase 3 verification violation. Pick a profile, then follow it.
- "I'll delegate the plan to a Plan subagent" → Phase 2 step 1 violation. Research with
Explore; plan inline. The dashboard can't surface delegated plans.
- "I'll skip
EnterPlanMode, plan mode resets bypassPermissions" → Phase 2 step 3 violation. The reset to default permission_mode is the accepted cost.
- "I'll just paste the plan as text instead of calling
ExitPlanMode" → Phase 2 step 4 violation. ExitPlanMode is the only acceptable submission.
- "The plan is obvious, let me start" → Phase 2 gate violation. Wait for approval.
- "Tests pass on my reading of the code" → no executable proof. Run the profile's command, or state why none applies.
- "I'll skip the worktree, it's a small change" → wrong skill. Use a feature branch directly without invoking this skill.
- "I'll bundle this unrelated cleanup into the feature commit" → split it. Open a separate PR.
- "User picked hand-off, but I'm already here — I'll just do Phase 3 myself" → exit cleanly. They opted out of inline implementation for a reason (context). Don't second-guess.
- "I'll write
.feature-plan-path later, after I start Phase 3" → write it now. /agent-dashboard:implement and resume can't find the plan without it.