adlc

name	adlc
description	Full Agent Development Life Cycle pipeline for the aiagent project. Runs story refinement → design → implementation planning (with human approval checkpoint) → implementation → testing → security → database → devops. Each coding phase uses a doer + antagonistic reviewer loop requiring score >= 8/10.
user-invocable	true
argument-hint	add a new tool that searches GitHub repositories
context	fork

ADLC Pipeline

Run the full development pipeline for:

---

Stage 1: Pre-Coding (no code is written until the plan is approved)

Step 1: Story Refinement (interactive)

Use the Agent tool with subagent_type: "story-refiner" to:

1. Explore the relevant aiagent codebase
2. Generate a focused batch of 3–6 clarifying questions
3. Present the questions to the user and collect answers
4. Repeat until the story is complete (all unknowns resolved)

Capture and store the full story-refiner output (both Refined Story and Relevant Codebase Context sections). This is passed verbatim to every subsequent agent.

When complete, ask: "Story looks complete — ready to design?"

---

Step 2: Design (review loop)

Use the Agent tool with subagent_type: "architect" to design the solution. Pass it:

• The full story-refiner output (Refined Story + Relevant Codebase Context)

The architect defaults to maintaining existing DDD patterns. Deviations are explicitly justified.

Then use the Agent tool with subagent_type: "architect-reviewer" to review the design. Pass the architect's full output.

Iterate up to 3 times until score >= 8/10 (APPROVED) or escalate to user.

Capture and store the approved design document — passed verbatim to the planner.

---

Step 3: Implementation Plan (review loop + human checkpoint)

Use the Agent tool with subagent_type: "planner" to produce an implementation plan. Pass it:

• The full story-refiner output
• The approved design document

The planner produces a Reuse Inventory and numbered implementation steps.

Then use the Agent tool with subagent_type: "planner-reviewer" to verify the plan.

Iterate up to 3 times until score >= 8/10.

PAUSE. Present the complete plan to the user (Reuse Inventory + all numbered steps).

Ask explicitly: "Does this plan look correct? Type 'proceed' to begin coding, or describe any changes needed."

Do NOT advance to Stage 2 until the user types 'proceed' or equivalent confirmation.

If the user requests changes, update the plan and re-present it.

Capture and store the approved plan — passed verbatim to the developer.

---

Stage 2: Coding (only after human approves the plan)

For each coding phase, pass to the agent: the approved plan (including Reuse Inventory), the approved design, and the refined story (including acceptance criteria).

---

Step 4: Implementation

Doer: Use the Agent tool with subagent_type: "developer"

Pass:

• Approved plan (Reuse Inventory + numbered steps)
• Approved design document
• Refined story (acceptance criteria)

The developer will:

• Establish a green test baseline first
• Execute one plan step at a time with test gates
• Run full QA when complete: go fmt ./... && go vet ./... && go mod tidy && go build . && go test ./... -race

Reviewer: Use the Agent tool with subagent_type: "developer-reviewer"

Iterate up to 3 times until score >= 8/10.

---

Step 5: Testing

Doer: Use the Agent tool with subagent_type: "tester"

Pass:

• Refined story (acceptance criteria — each criterion should map to at least one test)
• List of files changed by the developer

Reviewer: Use the Agent tool with subagent_type: "tester-reviewer"

Iterate up to 3 times until score >= 8/10.

---

Step 6: Security

Doer: Use the Agent tool with subagent_type: "security"

Pass:

• Refined story
• List of files changed

Reviewer: Use the Agent tool with subagent_type: "security-reviewer"

Iterate up to 3 times until score >= 8/10.

---

Step 7: Database (run if schema or repository files were changed)

Doer: Use the Agent tool with subagent_type: "dba"

Reviewer: Use the Agent tool with subagent_type: "dba-reviewer"

Iterate up to 3 times until score >= 8/10.

---

Step 8: DevOps (run if .github/workflows/, Dockerfile, or compose.yml were changed)

Doer: Use the Agent tool with subagent_type: "devops"

Reviewer: Use the Agent tool with subagent_type: "devops-reviewer"

Iterate up to 3 times until score >= 8/10.

---

Step 9: Acceptance Criteria Verification

After all coding phases complete, verify each acceptance criterion from the refined story:

For each criterion:

• Identify: which file and/or test satisfies it, and how
• If any criterion has no corresponding implementation or test: flag it as an open gap

Use the Agent tool with subagent_type: "developer" or subagent_type: "tester" passing the refined story and the full list of changed files.

---

Final ADLC Summary Report

Present a complete summary:

## ADLC Pipeline Complete

### Story
**Goal:** [one sentence]
**Acceptance Criteria:**
- [x] [criterion] — satisfied by [file/test]
- [ ] [criterion] — GAP: not implemented

### Design Summary
[Key design decisions, patterns reused vs. new]

### Reuse Inventory
| Component | File Path | Used in Step |
|-----------|-----------|-------------|

### Implementation Steps Completed
[numbered list with status: completed / skipped]

### Phase Scores
| Phase | Score | Iterations | Status |
|-------|-------|-----------|--------|
| Implementation | X.X/10 | N | APPROVED |
| Testing | X.X/10 | N | APPROVED |
| Security | X.X/10 | N | APPROVED |
| Database | X.X/10 | N | APPROVED / SKIPPED |
| DevOps | X.X/10 | N | APPROVED / SKIPPED |

**Overall Score:** X.X/10 (average of completed phases)

### Files Changed
[list of modified/created files]

### Remaining Issues
[Any warnings not yet addressed]

### Ready for PR: YES / NO (if NO, list blockers)

---

Important Rules

• Maximum 3 iterations per phase — if not approved after 3, present work with remaining issues flagged
• Human approval is required before Stage 2 — "proceed", "yes", "looks good" counts; silence does not
• Each doer agent CREATES/MODIFIES files; each reviewer only READS and SCORES
• The developer never skips test gates between steps
• All acceptance criteria must be verified before the summary report