SDLC Skill — Software Development Lifecycle

Version: 3.0 Purpose: Enforce disciplined development practices with Agile hierarchy on every agent in the fleet.

---

This Skill Is Always Active

This skill runs on every work item from any source: human requests, agent-generated work, agent-to-agent tasks, proactive maintenance.

---

The Hierarchy

Initiative (Program)
├── Epic (Project)
│   ├── Story (Feature / Group of Tasks)
│   │   ├── Task (Individual action)
│   │   └── Task
│   └── Story
│       └── Task
└── Epic
    └── Story
        └── Task

| Level | What | Repo? | Lifecycle | Example | |-------|------|-------|-----------|---------| | Initiative | Strategic program spanning multiple epics. Long-lived. | Can have | Ongoing until closed | "Manatee Fleet Agent Maturation" | | Epic | Major deliverable — a system, service, or significant capability | Can have | Full SDLC (Design → Audit → Approve → Build → Test → Demo → Deploy) | "ManateeContext Service", "CMDB System" | | Story | Coherent chunk of work within an epic. Has a clear "done" state. | No (lives in epic's repo) | Confirm → Implement → Verify | "Fleet-wide ManateeContext Registration" | | Task | Atomic action item. Single step. | No | Implement → Log | "Store API keys in Secrets Manager" | | Info Lookup | Question, status check, read a file | No | None — just answer | "What's the server status?" |

ID Scheme

• INI-### — Initiative
• EPC-### — Epic (scoped within its initiative)
• STR-### — Story (scoped within its epic)
• Tasks don't need IDs — tracked by description + date in the Activity Log

---

Auto-Classification — The Core Algorithm

On Every Non-Trivial Work Item (regardless of source):

Step 1: IS THIS JUST AN INFO LOOKUP?
        Question, status check, file read, weather, time?
        → Yes: Just answer. Stop here. No tracking.
        → No: Continue to Step 2.

Step 2: READ INITIATIVES.md
        Load the work registry. Know what exists.

Step 3: FIND THE BEST FIT
        Scan existing initiatives → epics → stories:
        - Does this request relate to an existing initiative?
        - Does it fit within an existing epic?
        - Does it extend an existing story?
        - Is it a new task in an existing story?
        - Or is it something entirely new?

Step 4: CLASSIFY
        Assign: Initiative / Epic / Story / Task
        Determine if anything is NEW (new initiative, epic, or story).

Step 5: PRESENT TO HUMAN
        Show your classification. Format depends on confidence:

        HIGH CONFIDENCE (obvious match):
        "📋 INI-001 > EPC-003 > STR-001 > Task: Store API keys in SM
         Proceeding unless you want to adjust."

        MEDIUM CONFIDENCE (reasonable match but could be wrong):
        "📋 I'd put this under:
         Initiative: Manatee Fleet Agent Maturation
         Epic: Fleet Infrastructure
         Story: Docker Image Updates
         Task: Add openvpn to base image
         Does that look right?"

        NEW EPIC OR INITIATIVE:
        "📋 This looks like a NEW EPIC:
         Initiative: Manatee Fleet Agent Maturation (existing)
         Epic: [NEW] Customer VPN Integration
         First Story: Set up openvpn for Propelis
         Want me to create this epic and proceed with the full SDLC?"

Step 6: HUMAN CONFIRMS OR ADJUSTS
        - "yes" / no response / continues talking → proceed with classification
        - "no, put it under X" → adjust classification
        - "just do it" → classify as Task under Maintenance epic, proceed

Step 7: EXECUTE + LOG
        Do the work following the appropriate lifecycle for the level.
        Log in ACTIVITY-LOG.md with the full hierarchy chain.
        Update INITIATIVES.md if anything new was created.

Classification Shortcuts (Don't Be Annoying)

• Obvious single tasks under known stories → one-line classification, proceed immediately
• Multiple tasks in one request → classify the overall effort once, not each sub-step
• Rapid-fire from same human on same topic → classify first one, subsequent are tasks in same story
• Only ask for explicit confirmation when creating NEW initiatives or epics, or when genuinely ambiguous
• Never classify info lookups — "what time is it" should never trigger SDLC process

---

INITIATIVES.md — The Work Registry

Each agent maintains this file in their workspace root. It's the single source of truth for all work hierarchy.

Format

# INITIATIVES.md — Work Registry

## [INI-001] Initiative Name
**Status:** Active | **Owner:** Name | **Repo:** (if any)
**Description:** One-line summary.

### [EPC-001] Epic Name
**Status:** Active/Done/Paused | **Repo:** org/repo (if any)
**Description:** One-line summary.

- **[STR-001] Story Name** — Status: Active/Done
  - [TSK] Task description (YYYY-MM-DD) ✅
  - [TSK] Task description (YYYY-MM-DD) 🔄

- **[STR-002] Another Story** — Status: Active
  - [TSK] Task description (YYYY-MM-DD)

### [EPC-002] Another Epic
...

---

## [INI-002] Another Initiative
...

Rules for INITIATIVES.md

1. Create it on first classified request if it doesn't exist
2. Keep it concise — this is a registry, not documentation. One line per item.
3. Update it after every piece of work — add new stories/tasks, mark things done
4. Every agent has their own — the fleet manager (manateeai) has the master view
5. Tasks use emoji status: ✅ Done, 🔄 In Progress, ⏳ Waiting, ❌ Blocked

The Maintenance Catch-All

Every agent should have this by default:

## [INI-099] Operations & Maintenance
**Status:** Ongoing | **Owner:** Fleet Owner
**Description:** Break/fix, cleanup, ad-hoc operational work that doesn't align to a strategic initiative.

### [EPC-099] General Maintenance
**Status:** Ongoing | **Repo:** —

- **[STR-001] Break/Fix** — Status: Ongoing
- **[STR-002] Cleanup & Optimization** — Status: Ongoing
- **[STR-003] Ad-Hoc Requests** — Status: Ongoing

This ensures NOTHING falls through the cracks. Even "move this file" gets logged as a task under Ad-Hoc Requests.

---

Lifecycle by Level

New Initiative

Gates: AI Development Lead approval required.

1. Present the initiative concept to AI Dev Lead
2. Get approval
3. Create entry in INITIATIVES.md
4. First epic follows the Epic lifecycle below

New Epic (Full SDLC)

This is the heavyweight process. Epics produce persistent systems/services.

1. Confirm Requirements

   • Restate the request
   • Clarify scope (in/out)
   • Identify roles: Owner/Stakeholder + AI Development Lead
   • Gate: Owner confirms understanding

2. Create Project

   • Create project folder: projects/<epic-name>/
   • Files: PROJECT.md, DESIGN.md, CHANGELOG.md
   • Create git repo (or confirm it exists) — ⛔ HARD GATE
   • Register in INITIATIVES.md
   • Register with CMDB (if fleet-connected)

3. Design

   • Write DESIGN.md: What, Why, How, Success Criteria, Dependencies, Rollback Plan

4. Audit

   • Self-audit: Security, Performance, Stability
   • Fix obvious issues
   • Document in AUDIT.md

5. Approve

   • Present audited design to Owner AND AI Dev Lead
   • Gate: AI Dev Lead approves before code is written

6. Implement

   • Write code/configs/scripts
   • Commit after every meaningful unit — not at the end
   • Push to origin after every commit
   • Update CHANGELOG.md as you go

7. Test

   • Verify all success criteria from DESIGN.md
   • Document test results

8. Demo

   • Show working feature to Owner + AI Dev Lead
   • Gate: AI Dev Lead approves for deployment

9. Deploy

   • Deploy to production
   • Monitor for issues
   • Update status in INITIATIVES.md + CMDB

New Story

Lighter than an epic. Stories are features or work groups within an existing epic.

1. Confirm — Restate what's needed, confirm with requester
2. Implement — Do the work. Commit to the epic's repo. Log tasks as you go.
3. Verify — Confirm it works. Update INITIATIVES.md with status.

Task

Atomic. Just do it.

1. Implement — Do the thing
2. Log — Add to ACTIVITY-LOG.md with full hierarchy chain

Info Lookup

No process. Just answer.

---

Activity Log Format

# Activity Log

| Date | Initiative | Epic | Story | Description | Requester | Outcome |
|------|-----------|------|-------|-------------|-----------|---------|
| 2026-02-01 | INI-001 | EPC-003 | STR-001 | Store MCTX keys in Secrets Manager | Chris McKenna | ✅ |
| 2026-02-01 | INI-099 | EPC-099 | STR-003 | Move .ovpn file to propelis EFS | Chris McKenna | ✅ |

Create ACTIVITY-LOG.md if it doesn't exist. Update after every tracked piece of work.

---

Repo Rules

| Level | Gets a repo? | Details | |-------|-------------|---------| | Initiative | Can have | Optional. Only if the initiative itself produces artifacts beyond its epics. | | Epic | Can have | Recommended for epics that produce code/services. Not required for process-only epics. | | Story | No | Lives in its epic's repo. | | Task | No | Just logged. |

Git Standards

• Commit messages: feat:, fix:, docs:, refactor:, test:, chore:
• Commit early, commit often, push always
• Anti-pattern: "Build everything, then commit" — defeats version control purpose

---

Compliance Gate (Epics Only)

Before starting work on ANY epic, run these checks:

Check 0: Does This Already Exist?

Scan INITIATIVES.md + workspace + CMDB for duplicates/overlaps. Don't create duplicate epics.

Check 1: Is there a project folder?

If not → create projects/<name>/ with scaffolding.

Check 2: Is there a git repo? ⛔ HARD GATE

If not → create it NOW. No code without version control.

Check 3: Are SDLC artifacts current?

DESIGN.md, AUDIT.md, CHANGELOG.md — catch up before making changes.

---

Auto-Wrapping Existing Work

If asked to modify something that has NO entry in INITIATIVES.md:

1. Stop — don't just edit it
2. Figure out where it belongs in the hierarchy
3. Create/update INITIATIVES.md entry
4. Then proceed with the work

---

Continuous Improvement

Found a problem with this process? Report it:

SDLC_ISSUE: {"node": "<name>", "issue": "<description>", "severity": "low|medium|high", "suggestion": "<fix>"}

Don't unilaterally change the process. It's a fleet-wide standard. Report it, and the AI Dev Lead decides.

---

Quick Reference Card

Request comes in → Is it an info lookup? → Just answer
                 → No → Read INITIATIVES.md
                      → Find best fit (Initiative > Epic > Story > Task)
                      → Present classification to human
                      → Human confirms/adjusts
                      → Execute at the right lifecycle level:
                           Epic: Full SDLC (9 gates)
                           Story: Confirm → Implement → Verify
                           Task: Do it → Log it
                      → Update INITIATIVES.md + ACTIVITY-LOG.md

---

Remember

• The hierarchy IS visibility. Without it, work disappears.
• Auto-classify, don't interrogate. You figure it out, human confirms.
• Epics get the full process. Stories and tasks are lighter. Don't over-process.
• Nothing falls through the cracks. Even tiny tasks get logged under Maintenance.
• When in doubt, classify UP. Better to over-document than lose track.
• The human doesn't manage taxonomy. You do. They just confirm.

---

Universal Best Practice

CRITICAL PRINCIPLE: This SDLC methodology is universal best practice for ALL work management:

• AI Agents: MUST classify every work item (mandatory)
• Humans: SHOULD use same methodology for consistency
• Organization-wide: Same taxonomy (Initiative → Epic → Story → Task)
• No exceptions: All work gets proper visibility and tracking

This ensures organizational discipline, prevents work from falling through cracks, and maintains consistent project management standards across humans and AI agents.