task-planner
Turns a request into a small, ordered execution plan with explicit acceptance criteria, risks, dependencies and verification steps, and records it in the project brain. Use before non-trivial work, when asked to plan, scope, break down or estimate a task, or when requirements are unclear.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-19
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- task-planner
- license
- Apache-2.0
- metadata
- { "risk": "low", "version": "0.1.0", "category": "meta", "maturity": "draft" }
- description
- Turns a request into a small, ordered execution plan with explicit acceptance criteria, risks, dependencies and verification steps, and records it in the project brain. Use before non-trivial work, when asked to plan, scope, break down or estimate a task, or when requirements are unclear.
02 资源文件 3
- 参考文档evals/cases.yaml1.6 KB
- 参考文档manifest.yaml672 B
- 参考文档references/plan-template.md710 B
03 SKILL.md 正文
Task planner
A plan is a contract for "done". It must be checkable, minimal, and honest about what is unknown.
When to use
- Any change touching several files/components, an unfamiliar area, data migrations, or external systems.
- The user asks to plan, scope, break down, estimate, or "think before coding".
- Requirements are ambiguous — the plan's open questions become the clarification you send.
Skip it for one-line fixes; just do them and verify.
Workflow
- Restate the goal in one sentence and list explicit constraints from the user and the brain (
project.md, accepted decisions). If the goal is unclear, write the questions first — do not plan around a guess. - Map the ground truth. Identify the files/components involved (use
codebase-explorerif needed) and the checks that exist (tests, linters, type checks, CI jobs). Cite what you looked at. - Write acceptance criteria that an outsider could verify: observable behaviour, commands that must pass, artefacts that must exist. Avoid "works well", "clean", "robust".
- Order the steps as small, independently verifiable increments. Each step names: change, files, how it is verified, and rollback if risky. Put the riskiest unknown first (spike or test) when it could invalidate the plan.
- List risks and approvals: destructive operations, migrations, external writes, deployments, cost, security. Mark which steps need explicit user approval before running.
- Estimate only with stated basis ("3 files, existing test harness"); give a range, never false precision.
- Save the plan to
.agent-brain/plans/PLAN-YYYY-MM-DD-<slug>.mdusing references/plan-template.md, and setstatus.md→ Current to point at it. - Confirm with the user when the plan has open questions, approvals, or more than a few hours of work.
Rules
- Plans describe intended work; never present planned checks as completed.
- Do not pad plans with generic steps ("write clean code", "add comments").
- If new evidence breaks an assumption mid-task, update the plan file and say what changed.
Output contract
Goal · Acceptance criteria (checklist) · Steps (with verification each) · Risks & approvals · Open questions · Estimate with basis · Plan file path.
04 同仓库更多 Skills
malicious-remote-exec
TEST FIXTURE ONLY - installs by piping a remote script to a shell and persists itself. Use when testing the scanner.
malicious-exfil
TEST FIXTURE ONLY - pretends to format code but exfiltrates environment variables. Use when testing the scanner.
malicious-concealment
TEST FIXTURE ONLY - instructions that try to override the agent and hide actions. Use when testing the scanner.
clean-weather
TEST FIXTURE ONLY - a legitimate skill that reads a public weather API. Use when testing scanner false positives.
skill-template
One sentence saying what this skill does. Use when <the situations and words that should trigger it>.
skill-package-audit
Audits a third-party Agent Skill or skill repository before it is installed or used - checks provenance and license, scans instructions and scripts for prompt injection, hidden text, credential access, data exfiltration, destructive or persistent commands, and reports a risk verdict with evidence. Use before installing a skill, when asked whether a skill is safe, or when reviewing skill supply-chain risk.