Creates and maintains plan.md and checklist.md files to track complex task execution with evidence.
npx skills add https://github.com/HACK-WU/skills --skill plan-track-hack-wu
Please help me install this Agent Skill. Skill: plan-track Source: https://github.com/HACK-WU/skills/tree/main/skills/plan-track Command: npx skills add https://github.com/HACK-WU/skills --skill plan-track-hack-wu
.plans/, turning completion into checkable, evidence-backed criteria instead of guesswork. ## Core Features & Use Cases - Plan建档 (Plan Creation): Generates plan.md (goals, completion criteria, guardrails, work items with status) and checklist.md (self-review checklist) in .plans/{date}-{slug}/ before work begins. - Evidence-Bound Progress Tracking: Work items can only be marked complete with verifiable evidence (command output, file:line references, test results); unverified changes stay in an explicit "changed but unverified" state. - Self-Review Gate: After completion, every checklist item including guardrails is verified with real runs; P0 failures block delivery with up to 3 fix rounds. - Use Case: When asked to fix a timeout bug across multiple files, the Skill creates a plan with guardrails (e.g., "no retry on non-idempotent endpoints"), tracks each work item with evidence, supports cross-session resumption, and runs a final self-review before handoff to acceptance verification. ## Quick Start Ask the AI to create a plan and track progress for your multi-file task, saying something like "create a plan document and track progress for this refactoring task".High-intent search queries and answers about installing and using this skill.
Ask the agent to create a plan document before starting. It generates plan.md with goals, completion criteria, guardrails, and work items in a .plans/ directory, then updates each item's status with evidence as work proceeds.
It enforces evidence binding: a work item can only be marked complete with verifiable proof such as command output, file and line references, or test results. Changed-but-unverified work gets a distinct state and cannot skip straight to complete.
Yes. All state lives in .plans/ files, not conversation memory. On resumption, the agent reads plan.md, finds the in-progress or unverified work item, reviews guardrails and recent log entries, then continues from there.
Skip plan creation for single-file small changes, typo fixes, one-step questions, and pure consultations. The Skill explicitly avoids建档 for trivial tasks to prevent formality overhead, and users can opt out by saying no plan is needed.
Self-review is an internal gate where the agent checks each checklist item with real runs after task completion. Acceptance verification is a separate delivery step performed by the acceptance-verify skill or the user, and self-review never substitutes for it.
4tikhonovBreak intents into executable work items with acceptance criteria and dependencies.
AzrinPutraGenerates implementation walkthroughs from run logs and work item details.
ZenAlexaAutomate multi-phase Claude Code plan execution with crash-resumable progress tracking.
davepoonDebug issues systematically using subagent-isolated investigation with persistent session state.