skills/satone7/skills/executing-single-task

executing-single-task

SKILL.md

Executing Single Task

Overview

This skill is the execution counterpart to find-next-task. It takes a single task (via the full find-next-task output JSON), executes it with minimal context, verifies results with evidence, updates the plan JSON by setting only passes, and commits once.

Announce at start: "I'm using the executing-single-task skill to execute exactly one task and update passes."

Required First Step

Invoke using-superpowers before any response or action. If any additional skill might apply (even 1%), invoke it before proceeding.

Input

The input is the full JSON output of find-next-task (verbatim). Treat plan_file as the source of truth.

Fail-fast conditions (no file changes):

  • error != null
  • selection_required == true
  • next_task == null
  • plan_file missing or not an absolute path
  • next_task.id missing

Output

Output exactly one JSON object and nothing else (no prose, no markdown fences).

{
  "plan_file": "/abs/path/to/plan.json",
  "task_id": 4,
  "result": "SUCCESS",
  "passes_written": true,
  "commit": {
    "created": true,
    "sha": "abc123",
    "message": "feat(task): 4 Login Page Implementation"
  },
  "verification_evidence": [
    {
      "kind": "command",
      "command": "npm test",
      "exit_code": 0,
      "summary": "PASS (all tests)"
    }
  ],
  "errors": []
}

result values:

  • SUCCESS: Task executed, verified, passes set to true, commit created.
  • FAIL: Task executed but verification failed or incomplete, passes kept/set to false, commit may or may not exist (see rules).
  • BLOCKED: Cannot proceed (invalid input, missing task, parse failure). No changes made.

Workflow

1. Validate input JSON

If fail-fast conditions match, output result: "BLOCKED" with an errors[] entry and stop.

2. Load plan and locate task

  1. Read plan_file and parse JSON.
  2. Locate the task whose id matches next_task.id.
  3. If not found, output result: "BLOCKED" and stop.

3. Execute steps with strict boundaries

Execute task.steps in order.

Boundaries:

  • Prefer task.files.create/modify/test as the only files you read/modify.
  • Do not perform broad repo exploration unless a step explicitly requires it.
  • If a step is ambiguous, invoke the most relevant skill before acting.

4. Verification and pass/fail decision

You may set passes: true ONLY with fresh, evidence-backed verification.

Verification priority:

  1. Explicit verification commands in task.steps
  2. Evidence-backed checks derived from validation_criteria (if present)
  3. If still ambiguous: invoke using-superpowers, then invoke the most relevant verification skill(s) and follow them to obtain verification evidence

Record each verification as an entry in verification_evidence.

5. Update plan JSON (passes only)

Update only the located task’s passes field in plan_file.

  • Write passes: true only when verification evidence supports success.
  • Otherwise write or keep passes: false.
  • Do not add/remove/reorder any other fields.

6. Git commit (exactly one)

Create exactly one commit for this task.

Commit must include:

  • Any code changes made for the task
  • The plan JSON change (the updated passes)

Commit message:

  • feat(task): <id> <title> when implementing functionality
  • fix(task): <id> <title> when fixing a bug
  • Otherwise: chore(task): <id> <title>

If result: "BLOCKED", do not commit.

7. Emit final JSON output

Return the JSON object matching the schema above. No additional text.

Weekly Installs
3
Repository
satone7/skills
First Seen
5 days ago
Installed on
trae3
claude-code3
github-copilot3
codex3
kimi-cli3
gemini-cli3