Artifact numbering

Artifact Numbering Skill v1.0

The Core Rule

NEVER use artifact numbers from a plan file or memory. Numbers in plans decay. Live canonical documents are the only authoritative source.

At the start of every Phase 9+ session:

# Read live canonical docs — do not use any cached or plan-file numbers
grep "Factory:\|F-[0-9]" AGENTS.md | tail -5
grep "Task Type:\|T-[0-9]" AGENTS.md | tail -5
grep "Skill:\|SK-[0-9]" AGENTS.md | tail -5

---

When to Invoke

• When assigning a new factory ID (F-XXXX)
• When assigning a new task type number (T-XXX)
• When assigning a new skill number (SK-XXX)
• When assigning a new BFA rule or family ID
• Before any STATE-Pn.json is saved with artifact numbers

---

Numbering Protocol

Step 1: Read live numbers (MANDATORY at session start)

# From AGENTS.md or ENGINE_ARCHITECTURE_MERGED:
grep -i "next.*factory\|last.*factory\|F-[0-9][0-9][0-9][0-9]" AGENTS.md | tail -3
grep -i "next.*task\|last.*task\|T-[0-9][0-9][0-9]" AGENTS.md | tail -3
grep -i "next.*skill\|last.*skill\|SK-[0-9][0-9][0-9]" AGENTS.md | tail -3

Record in STATE-Pn.json:

{
  "artifactNumbers": {
    "readAt": "session-start",
    "nextFactory": "F-XXXX (from AGENTS.md)",
    "nextTaskType": "T-XXX (from AGENTS.md)",
    "nextSkillNumber": "SK-XXX (from AGENTS.md)"
  }
}

Step 2: Cross-check ENGINE_ARCHITECTURE_MERGED

Before claiming a number, verify it is not already in use:

grep "F-1340\|F-1341" server/src/engine-contracts/ -r
grep "T-517\|T-518" server/src/engine-contracts/ -r

Step 3: Assign sequentially

• Factory IDs: next sequential integer after the highest in use
• Task types: next sequential integer after the highest in use
• Skill numbers: next sequential integer after the highest SK-XXX registered in agent-constitution/SKILL.md
• Never skip numbers — gaps create audit confusion
• Never reuse a decommissioned number — use the next available

Step 4: Record in STATE-Pn.json

{
  "reservedThisSession": {
    "factories": ["F-XXXX"],
    "taskTypes": ["T-XXX"],
    "skillNumbers": []
  }
}

This prevents a second Codex session from claiming the same numbers before the first session's commit is pushed.

---

Collision Prevention

Collision type	Prevention
Two sessions assign same F-XXXX	Both check STATE-Pn.json at session start; higher session uses next available
Plan file says F-1339 but live AGENTS.md says F-1342	Live doc wins; plan file number is stale
Skill number already assigned in agent-constitution	Read skill registry before assigning; always use next available
Task type T-516 assigned but contract not yet committed	Reserve in STATE-Pn.json; don't reuse until commit confirmed

---

Numbers at Plan Creation (REFERENCE ONLY — DO NOT USE)

These numbers are from AGENTS.md at plan creation time. By the time you read this, they may be stale. Read live canonical docs.

F-1339 (plan creation) → READ LIVE before assigning next F-XXXX
T-516  (plan creation) → READ LIVE before assigning next T-XXX
SK-330 (plan creation) → READ LIVE before assigning next SK-XXX

---

Anti-Patterns

1. "The plan says the next factory is F-1340." Plan numbers decay. Read live docs.
2. "I'll assign the number and update the docs later." Assign only after reading live. Record in STATE-Pn.json immediately.
3. "Only one session runs at a time." Multiple sessions can run in different worktrees. STATE-Pn.json reservation prevents collision.
4. "This skill number is unused — I'll reclaim it." Never reuse a number. Use next sequential.

---

Universal Bits (UUS G07) — live-doc-wins, named canonical source for mvp, reserve in STATE

These are the universal cross-project bits this skill must carry (imported from core via the universal-skills mapping), TS-adapted for the mvp stack.

Live doc wins — never number from plan or memory (universal)

The portable rule: never take an artifact number from a plan file or memory. Before assigning, read the live canonical document for the highest number in use, cross-check that the candidate is not already used, and reserve it in STATE immediately so a parallel session/worktree cannot claim the same number before commit. Never reuse a deleted/decommissioned number — always next sequential.

Named canonical source for mvp (this is the part the inherited core file leaves blank)

For mvp, "live canonical docs" are not DECISIONS-LOCKED.md / SUBPROJECT-MAP.md (those are core). The mvp live sources are:

• DOCUMENT_INDEX.md + the project STATE-JSON guides (GUIDE-B0x-*-STATE-JSON family) — the authoritative registry of current numbers.
• Version numbers like GUIDE-Bxx-…-vN live inside the file names; the highest live vN is read from the actual file list, not from a plan.

# Read live highest numbers from the mvp index (not from a plan):
grep -nE "D-[0-9]+|BUG-[0-9]+|SK-[0-9]+" DOCUMENT_INDEX.md | tail -10
ls GUIDE-B*-*-v*.md | sort -V | tail -5     # highest live guide version

mvp has many versioned guides/plans (v9.1 / v10 / v30…v32) and competing indexes, so collisions and stale numbers are most likely here — that is exactly why the canonical live source (DOCUMENT_INDEX.md + STATE-JSON guides) must be named, so "live doc wins" has a concrete anchor. Record every assigned number in the project STATE-JSON the moment it is reserved.