Skip to content
Back to skills

Execute Feat

ASecurity

Execute one feature (FEAT-XXX) at a time using docs/forge/ideas/<IDEA_ID>/latest/tasks.md as the source of truth. Creates a short workspace checklist and tracks progress so reruns continue automatically.

  • 666 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added February 8, 2026
developmentgogit

Works with

  • mcp

Security analysis

A100/100

Scanned February 10, 2026

npx -y skills add majiayu000/claude-skill-registry --skill execute-feat --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Execute Feat?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Execute Feat
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-execute-feat/badge)](https://www.skillsdirectory.com/skills/majiayu000-execute-feat)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: Execute Feature
description: Execute one feature (FEAT-XXX) at a time using docs/forge/ideas/<IDEA_ID>/latest/tasks.md as the source of truth. Creates a short workspace checklist and tracks progress so reruns continue automatically.
argument-hint: "<IDEA_ID> [FEAT-XXX]   (examples: IDEA-0001-my-idea | IDEA-0001-my-idea FEAT-003 | FEAT-003)"
disable-model-invocation: true
---

# Feature Execution Assistant (Idea-Scoped, Tasks-Driven)

Execute development iteratively using **Features (FEAT-XXX)** as the unit of work.

This command must NOT depend on `latest/work_packages.md`.
The canonical source of truth is the **feature-grouped tasks** in:

- `docs/forge/ideas/<IDEA_ID>/latest/tasks.md`

Supports optional arguments:

- `/work-feat IDEA-0002-my-idea` -> auto-pick the next not-yet-completed FEAT from tasks.md using the progress tracker
- `/work-feat IDEA-0002-my-idea FEAT-003` -> execute a specific feature
- `/work-feat FEAT-003` -> best-effort: locate the FEAT by scanning idea tasks (only if unique)

Use `$ARGUMENTS` for IDEA_ID and/or FEAT id.

---

## Resolve IDEA_ID (required)

Before using any idea-scoped paths:

- If `$ARGUMENTS` starts with `IDEA-`, treat the first token as `IDEA_REF`.
- Call `vf.resolve_idea_id` with `idea_ref = IDEA_REF` and store `IDEA_ID`.
- Use `IDEA_ID` for all paths, YAML headers, and run log entries.

---

## Inputs (Auto)

### Canonical sources of truth (idea-scoped)

- `docs/forge/ideas/<IDEA_ID>/latest/tasks.md` (required; feature-grouped task definitions)
- `docs/forge/ideas/<IDEA_ID>/latest/existing_solution_map.md` (context for existing code, update when necessary)
- `docs/forge/ideas/<IDEA_ID>/latest/idea_normalized.md` (broader project context)

### Optional context

- `docs/forge/ideas/<IDEA_ID>/latest/features_backlog.md` (fallback to features.md if missing)
- `docs/forge/ideas/<IDEA_ID>/latest/epics_backlog.md` (fallback to epics.md if missing)
- `docs/forge/ideas/<IDEA_ID>/latest/concept_summary.md`
- `docs/forge/ideas/<IDEA_ID>/latest/existing_solution_map.md`
- `docs/forge/ideas/<IDEA_ID>/latest/codebase_context.md`
- `docs/forge/ideas/<IDEA_ID>/inputs/idea.md`

### Execution tracking (small, per-idea)

- `docs/forge/ideas/<IDEA_ID>/latest/feature_execution_progress.md` (create if missing; append-only)
- `docs/forge/ideas/<IDEA_ID>/planning/` (workspace checklists live here)
- `docs/forge/ideas/<IDEA_ID>/manifest.md` (optional best-effort)
- `docs/forge/ideas/<IDEA_ID>/run_log.md` (append-only)

---

## Step 0 - Safety & Repo Context

1) Read current repo state:
- `git status -sb`
- `git diff --stat`

2) If there are unrelated uncommitted changes:
- Ask the user to confirm whether to stash/commit before proceeding.
- Do NOT proceed blindly.

---

## Step 1 - Determine target FEAT

### Accepted invocation forms

**Form A:** `$ARGUMENTS` starts with `IDEA-...`
- Parse:
  - `IDEA_REF = first token`
  - Optional `FEAT_ID = second token` if it matches `FEAT-\d{3}`
  - Resolve `IDEA_ID` using `vf.resolve_idea_id`

**Form B:** `$ARGUMENTS` is only `FEAT-XXX`
- Scan all idea folders under `docs/forge/ideas/` for a `latest/tasks.md` containing that FEAT header.
- If exactly one match is found, set `IDEA_ID` accordingly.
- If zero matches: STOP and report "FEAT not found in any idea."
- If multiple matches: STOP and list candidates; ask the user to rerun with explicit `IDEA_ID`.

If no `FEAT_ID` was provided:
- Auto-select the next FEAT not marked complete in `latest/feature_execution_progress.md`.

### IMPORTANT: Do not scan the entire tasks.md
When selecting or executing a feature:
- Use targeted search within `latest/tasks.md` to locate the FEAT section header (e.g., `FEAT-003:` or `## FEAT-003`).
- Extract ONLY the tasks under that FEAT section.

If tasks.md format varies, adapt by finding a stable "feature header" pattern used in the file and extracting the block until the next feature header.

---

## Step 2 - Confirm required files exist

Required:
- `docs/forge/ideas/<IDEA_ID>/latest/tasks.md`

Create if missing:
- `docs/forge/ideas/<IDEA_ID>/latest/feature_execution_progress.md`
- `docs/forge/ideas/<IDEA_ID>/planning/`

If `latest/tasks.md` is missing, STOP and report the expected path.

---

## Step 3 - Build the Feature Workspace Checklist (short)

Create (or update) a checklist doc for this feature:

- `docs/forge/ideas/<IDEA_ID>/planning/FEC-<FEAT_ID>-workspace-checklist.md`

Checklist requirements (keep it short):
- Title: `<FEAT_ID> - <feature title>`
- Idea-ID and Run-ID
- A checkbox list of the tasks in this feature (TASK-### ids + titles)
- For each task: 1-2 bullets:
  - key files expected to change (from existing_solution_map if available; otherwise best-effort)
  - verify command(s) (best-effort)
- "Notes / Decisions" section (append-only)

Do NOT create long plan docs. This is a working checklist.

---

## Step 4 - Mark Feature as In Progress (progress tracker)

Update `docs/forge/ideas/<IDEA_ID>/latest/feature_execution_progress.md`:
- If this FEAT is not present, add an entry:
  - `FEAT-XXX - In Progress - <timestamp> - run <RUN_ID> - checklist: planning/FEC-FEAT-XXX-workspace-checklist.md`
- If already present and In Progress, treat as resume.

---

## Step 5 - Execute Tasks One at a Time (Iterative Loop)

For each TASK in the feature's section (in dependency order if stated; otherwise file order):

### 5.1 - Preparation
- Use `existing_solution_map.md` if present to prefer extension of existing code.
- Do minimal, focused changes.
- Keep commits small; include `TASK-###` in commit messages.

### 5.2 - Implement
- Implement the smallest change meeting acceptance criteria.
- Avoid parallel implementations; extend existing components where possible.

### 5.3 - Verify
- Run the most relevant verification (unit tests/lint/smoke).
- If failing: attempt focused fix; re-run.

### 5.4 - Update checklist
In `planning/FEC-...-workspace-checklist.md`:
- Check off the task only after verification passes.
- Add 1-3 bullets:
  - files changed
  - commands run
  - anything learned

### 5.5 - If blocked
- Record blocker under that task in the checklist
- Update progress tracker to `Blocked` for this FEAT
- STOP

---

## Step 6 - Feature-level Close Out

After all tasks are checked:
1) Run a feature-level verification (best-effort):
   - run the main test suite subset relevant to changed areas
2) If passes:
   - Update progress tracker: `FEAT-XXX - Done - <timestamp> - run <RUN_ID>`
3) If fails:
   - Keep FEAT as In Progress
   - Record failure summary in checklist and STOP

---

## Step 7 - End-of-Run Summary (Always)

Print a concise summary:
- Feature executed: FEAT-XXX
- Tasks done vs remaining
- Verification commands run + result
- Key files touched
- Next recommended feature (next not-done FEAT)

---

## Required tool calls (MCP helpers)

- `vf.start_run` (idea_id=<IDEA_ID>, label="work-feat")
- `vf.write` to create/update:
  - `planning/FEC-<FEAT_ID>-workspace-checklist.md`
  - `latest/feature_execution_progress.md`
  - run snapshots under `runs/<RUN_ID>/...` (best-effort)
- `vf.append_log` with stage `execute.feature` and outputs

---

## Non-negotiable Rules

- Source of truth is `latest/tasks.md`.
- Execute exactly one feature per invocation.
- Do not read the full tasks.md; extract only the FEAT section being executed.
- Prefer extending existing code over creating parallel subsystems (use existing_solution_map when available).
- Only mark tasks done after verification passes.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.