run-agent-task

name	run-agent-task
description	Thin Claude adapter for spec-harness implement-task on one .spec feature TASK.

run-agent-task

Use this adapter only to execute one canonical .spec/<feature-name> TASK.

Required input:

• Feature: <feature-name>
• Task: <TASK-ID>

Canonical mapping:

spec-harness
Mode: implement-task
Feature: <feature-name>
Task: <TASK-ID>

Required behavior:

1. Read AGENTS.md.
2. Read .agents/skills/spec-harness/SKILL.md.
3. Read the feature SPEC, SDD, TASKS, AGENT_RULES, task-scope, acceptance file, and prompts/implement-task.md.
4. Modify only files allowed by the current TASK scope.
5. Run the required Harness checks and TASK-specific checks.
6. Write the canonical Markdown report and evidence JSON under .spec/<feature-name>/reports/.

Do not create provider-private task state, branch policy, completion rules, or review verdicts. Do not commit, push, merge, or update local Claude permissions.