tdd-slice

Files (4)

• SKILL.md
• session.md · examples/
• failure-report-format.md · references/
• retry-logic.md · references/

SKILL.md (first lines)

Implement a complete slice by orchestrating multiple micro-tdd-agent cycles, validating each cycle, and aggregating results.

## Subagents Used

This command delegates work to subagents:
- `micro-tdd-agent` - for test behaviors
- `micro-refactor-agent` - for refactorings
- `commit-agent`
- `tdd-validation-agent`
- `micro-fix-agent` - for fixing trivial validation issues
- `investigator-agent` - for analyzing repeated validation failures

## What This Command Does

**Input**: Slice requirements document containing:
- Slice number and name
- Slice goal
- Features to implement
- Tests to write (list of test behaviors)
- Commit message (for reference, not used - micro commits kept)

**Output**: Path to tdd-agent report referencing all micro and validation reports

**Does**:
- Analyze requirements and create execution plan
- Call micro-tdd-agent for test behaviors
- Call micro-refactor-agent for refactorings
- Call commit-agent to create commits for each cycle
- Validate each commit with tdd-validation-agent
- Attempt to fix trivial validation issues (commit message, comments, one-liners) with micro-fix-agent
- Retry failed steps with context (full reset if fixes fail or issues are substantial)
- Call investigator-agent after second validation failure to determine root cause
- Analyze and report on repeated failures with investigation findings
- Create summary report with references to child reports

**Does NOT**:
- Write tests or code directly (delegates to agents)
- Squash commits (keeps individual micro commits)
- Run cleanup (micro agents handle that)

## Input Format

Slice requirements should be provided as:

```markdown
# Slice [N]: [Slice Name]

## Goal
[What this slice achieves]

## Features
- Feature 1
- Feature 2

## Tests to Write
1. Test behavior 1 description
2. Refactor: [refactoring description]
3. Test behavior 2 description

## Commit Message
[Reference message - not used, micro commits kept]
```

**Note**: Items may be prefixed with "Refactor:" to explicitly indicate refactorings, but the agent will analyze each item to determine if it's a behavioral change (test) or structural change (refactor).

## Workflow

### Phase 1: Analyze & Plan

**Step 1: Read slice requirements**
- Parse slice number, name, goal
- Extract list of test behaviors
- Note features to implement

Post "✅ Slice requirements parsed: [slice name]"

**Step 2: Analyze codebase**
- Examine relevant test files and production code
- Understand existing patterns and conventions
- Identify dependencies and structure

Post "✅ Codebase analyzed"

**Step 3: Create execution plan**
- Review items from "Tests to Write" section
- For each item, analyze whether it's a behavioral change or structural change:
  - **Test behavior** (use micro-tdd-agent): Adds new functionality, changes behavior, adds assertions about system behavior
  - **Refactoring** (use micro-refactor-agent): Improves code structure without changing behavior (extract method, rename, reorganize)
  - Items explicitly prefixed with "Refactor:" are refactorings, but also identify refactorings without the prefix
- Ensure test behaviors are specific and atomic (one behavior per test)
- Ensure refactorings are specific and atomic (one structural change)
- If items are too broad, break them down further
- Create ordered list with type annotations (test/refactor) and descriptions

Post "✅ Execution plan created: [N] items ([X] tests, [Y] refactors)"

### Phase 2: Execute Micro Cycles

For each item in the execution plan:

**Step 1: Call appropriate agent**

Based on the execution plan's type annotation for this item:

**If item is a test behavior:**
Delegate to `micro-tdd-agent` with the test behavior description.

**If item is a refactoring:**
Delegate to `micro-refactor-agent` with the refactoring description (remove "Refactor:" prefix if present).

Track:
- Attempt number (1 or 2)
- Agent type used (micro-tdd or micro-refactor)
- Micro report path
- Commit hash created

**If agent fails:** Read `references/retry-logic.md` ("Micro Agent Retry" section) and follow it. Do not improvise.

**If agent succeeds:**

From MihaiOnSoftware/.ai, MIT.

More from MihaiOnSoftware/.ai