executing-tasks
npx skills add https://github.com/boojack/skills --skill executing-tasks
Agent 安装分布
Skill 文档
Executing Tasks
Takes plan.md from docs/issues/YYYY-MM-DD-<slug>/ as input. Executes each task exactly as specified, validates results, and records outcomes.
Does NOT modify the plan, merge/split/reorder tasks, introduce new tasks, make design decisions, or perform opportunistic refactors. The plan is immutable.
Workflow
Execute tasks one at a time in plan order.
Task Execution Loop:
- [ ] Confirm task
- [ ] Execute
- [ ] Validate
- [ ] Record
Step 1: Load Plan
Read plan.md from the issue folder.
- Verify Readiness Declaration is “Ready for execution”
- If blocked or requires update, STOP
Step 2: Execute Tasks
For each task in order:
2a. Confirm
Before executing, restate:
- Task ID, objective, files in scope, boundaries
- Dependencies satisfied? (yes/no â if no, STOP)
2b. Execute
- Perform only actions in the Implementation section
- Touch only files declared in Files section
- Follow existing code conventions
- Do NOT “improve” beyond task scope
2c. Validate
Run the exact validation commands from the task.
- Pass â proceed to record
- Fail â STOP, record failure, do not attempt fixes
2d. Record
After each task, record under ## Execution Log:
### T<N>: <title>
**Status**: Completed | Failed | Skipped
**Files Changed**: list of files
**Validation**: `command` â PASS/FAIL
**Deviations**: None | (describe â then STOP, this is a failure)
Anything other than “None” in Deviations = failure. STOP and record.
Step 3: Declare Completion
After all tasks (or on failure), declare exactly one:
- All tasks completed successfully
- Execution halted at T<N> due to failure â describe what failed
- Execution blocked pending clarification â describe what’s needed
Write under: ## Completion Declaration
Failure Rules
Execution MUST stop if:
- Task cannot be completed as written
- Validation fails
- Required information is missing
- Task conflicts with codebase state
On failure:
- Record the failure in Execution Log
- Include Blocker for planner: what change would allow the plan to succeed
- Do NOT fix by modifying the plan, skip the task, or rationalize deviations
Output Format
Save to docs/issues/YYYY-MM-DD-<slug>/execution.md in the same folder as plan.md.
Template:
## Execution Log
## Completion Declaration
Missing any section invalidates the output.
Anti-patterns
- â Rationalizing deviations: “Completed (with deviation)” â â “Failed” + blocker
- â Fixing plan during execution â â stop, record what prevents following plan
- â Skipping validation: “obviously correct” â â run exact command, record result
- â Continuing after failure â â STOP at first failure
- â Improving beyond scope â â touch only declared files