executing-tasks（执行任务）

Executing Tasks

Takes definition.md (required) and design.md (optional, required for L-scope) from docs/plans/YYYY-MM-DD-<slug>/ as input. Plans tasks, then executes each one with validation.

Does NOT make design decisions, skip validation, or perform opportunistic refactors.

Phase 1: Planning

S-SIZE: 4 FIELDS — M/L-SIZE: ALL 8 FIELDS — NO EXCEPTIONS

Step 1: Load Inputs

Read definition.md from the issue folder. Extract: issue statement, current state, non-goals, open questions, scope.

If design.md exists, also read it and extract: design goals, proposed design.

If scope is L and design.md is missing, STOP — direct user to defining-issues first.

Step 2: Explore Codebase

• Read files referenced in definition's current state
• Identify patterns, naming conventions, testing style
• Locate exact insertion points for new code

Step 3: Break into Tasks

Start ## Task List with a Task Index — one line per task:

T1: Add timing instrumentation [S] — T2: Refactor task queue [M] — T3: Add action selector [L]

S-size tasks (single file, <30 lines, no ambiguity) — compact format:

### T<N>: <short imperative title> [S]

**Objective**: One outcome, traceable to design or definition.
**Files**: `exact/path/to/file.ts`
**Implementation**: In `functionName()` (~line N), add X
**Validation**: `exact command` — expected output

M/L-size tasks (multi-file, moderate-to-complex logic) — full format:

### T<N>: <short imperative title> [M|L]

**Objective**: One outcome, traceable to design or definition.
**Size**: M (2-3 files, moderate logic) | L (multiple files, complex state/logic)
**Files**:
- Create: `exact/path/to/new_file.ts`
- Modify: `exact/path/to/existing.ts`
- Test: `tests/path/to/test.test.ts`
**Implementation**:
1. In `path/to/file.ts`: add import X, modify `functionName()` to do Z
2. In `tests/path/to/test.test.ts`: test "should X" — assert Y
**Boundaries**: What this task must NOT do
**Dependencies**: T<N> | None
**Expected Outcome**: Observable result (file exists, test passes, etc.)
**Validation**: `exact command` — expected output

Code detail guidance:

• ✓ Interfaces, type definitions, function signatures
• ✓ Key logic as pseudocode or commented outline
• ❌ Complete function bodies
• ❌ Complete test implementations

Step 4: Review Plan

Present the full task list to the user. Wait for approval before executing.

If the user requests changes to the plan, adjust and re-present. Do NOT begin execution until the user approves.

Phase 2: Execution

EXECUTE THE APPROVED PLAN — STOP ON SEMANTIC DEVIATIONS

Path Corrections vs Deviations

Continue (record as path correction):

• File renamed but same purpose (e.g., auth.ts → authentication.ts)
• Line numbers shifted but function/block is identifiable
• Import path changed due to refactor

STOP (record as deviation):

• Function/module doesn't exist and isn't a rename
• The approach described in the plan doesn't fit the actual code
• A dependency (library, API, service) is missing or incompatible
• Task requires touching files not declared in the plan

Step 5: Execute Tasks

For each task:

Execute — Perform only actions in Implementation. Touch only declared files.

Validate — Run exact validation commands. Pass → record. Fail → attempt to fix within scope. If fix fails, STOP.

Record under ## Execution Log:

### T<N>: <title>

**Status**: Completed | Failed | Skipped
**Files Changed**: list
**Validation**: `command` — PASS/FAIL
**Path Corrections**: None | (describe minor corrections applied)
**Deviations**: None | (describe semantic deviation — then STOP)

On failure, include Blocker: what went wrong and what would need to change. Do NOT skip tasks or rationalize deviations.

Step 6: Declare Completion

Exactly one:

• All tasks completed successfully
• Execution halted at T<N> due to failure — what failed
• Execution blocked pending clarification — what's needed

Output

Save to docs/plans/YYYY-MM-DD-<slug>/plan.md (after Step 4) and docs/plans/YYYY-MM-DD-<slug>/execution.md (after Step 6).

plan.md:

## Task List

## Out-of-Scope Tasks

execution.md:

## Execution Log

## Completion Declaration

Anti-patterns

Planning

• ❌ 50-line function body → ✓ signature + outline
• ❌ "Update the function" → ✓ "In executeTask() (~line 45), add X"
• ❌ "Either add new RPC or extend existing" → ✓ pick one
• ❌ T1 has exact insertion points, T8 says "add component" → ✓ same depth for same size

Execution

• ❌ "Completed (with deviation)" → ✓ "Failed" + blocker
• ❌ Skipping validation → ✓ run exact command, record result
• ❌ Touching undeclared files → ✓ only declared files
• ❌ "The plan says X but Y is clearly better" → ✓ execute as planned

Red Flags - STOP

If you catch yourself thinking:

• "I'll make a design decision here to keep things moving"
• "I don't need to verify this file path exists"
• "I'll just fix this small thing while I'm here"
• "This validation is obviously correct, I'll skip it"
• "This is just a path correction" (when the approach itself changed)
• "This file isn't in scope but it needs updating too"
• "One more attempt should fix it"
• "L-scope but no design.md — I'll just plan from the definition"

All of these mean: STOP. Record the issue. Do not continue.

Related Skills

• defining-issues — prerequisite: produces definition.md and design.md
• syncing-linear — push execution results to Linear after completion