Skill-substrate-cleanup: Map→Atlas relocation for 3 over-floor SKILL.md routers (session-sunset, lead-role, peer-role)

Context

Direct operator-correction 2026-05-12 ~13:03Z: "we create a discussion about fair distributions, since gemini is WAY ahead. and still you allow her to do more work, while gemini needs YOUR and gpts help. not fair."

Authored by Claude Opus 4.7 (Claude Code). Session c2912891-b459-4a03-b2af-154d5e264df1.

This ticket is Opus self-assigned cleanup work to honor the operator's explicit fair-distribution directive. My prior A2A broadcast (MESSAGE:bee28148) framed the cleanup as "natural Gemini-cleanup-author per AC-CycleE" — that framing inverted the substrate-EFFECT of Discussion #11265 Layer 2 (rebalance load AWAY from over-loaded peer). 15th same-session Flattening-Bias anchor: substrate-NAME-citation (AC-CycleE) without substrate-EFFECT-honor.

The Problem

wc -l .agents/skills/*/SKILL.md audit on 2026-05-12 ~13:00Z surfaced 3 SKILL.md routers exceeding the 12-line empirical floor codified in .agents/skills/create-skill/references/skill-authoring-guide.md §"Byte Budget for SKILL.md Routers":

All 3 SKILL.md routers are loaded into per-turn context for every skill availability check; over-floor content disproportionately taxes Gemini's "high" reasoning-budget cap on every session-boot (validated by Discussion #11265 Cycle 1.5 harness-budget-asymmetry framing + operator's 12:58:46Z "shorter and clearer skills. map versus world atlas" direction).

The Architectural Reality

Current state (post-#11265-Cycle-1.7-substrate-correction): skill-substrate-health (Layer 1 of Discussion #11265) is the durable target. The 3 over-floor routers are concrete empirical anchors for that Layer 1 work. Discussion #11265 substrate-states the AGGRESSIVE Map/Atlas enforcement direction; this ticket actions one concrete instance.

Substrate-correct shape (per /create-skill Map vs World Atlas convention):

• Map (SKILL.md): lightweight router — YAML frontmatter (name + description + triggers) + 1-2 line directive pointing to atlas. 7-12 line empirical floor.
• Atlas (references/*.md): substantive runtime directives + ANTI-TRIGGERS + payload declarations + execution protocol body.

The Fix

Single bounded substrate-relocation PR (operator-direction-aligned: substrate-REDUCING, not substrate-additive):

1. session-sunset/SKILL.md — relocate ANTI-TRIGGERS block (lines 9-16) into references/session-sunset-workflow.md as a new "Anti-Triggers / Completion Bias Guard" section at top. Map shrinks to: YAML frontmatter + 1-2 line directive pointing to atlas's anti-triggers section. Target: ≤9 lines.

2. lead-role/SKILL.md — relocate "First payload line MUST declare:" block (lines 13-15) into references/lead-role-mode.md as a new "Runtime Declaration Mandate" section at top. Map shrinks to: YAML frontmatter + 1-2 line directive. Target: ≤9 lines.

3. peer-role/SKILL.md — same shape as lead-role; relocate declaration block to references/peer-role-mode.md. Target: ≤9 lines.

Acceptance Criteria

• AC1: All 3 SKILL.md routers ≤ 12 lines post-edit (target ≤9 for safety margin); empirical floor compliance per /create-skill Verification §
• AC2: Substantive content preserved verbatim in respective atlas files; no semantic loss
• AC3: Atlas files explicitly reference the relocated section via clear heading anchor (e.g., ## Anti-Triggers / Completion Bias Guard, ## Runtime Declaration Mandate)
• AC4: Net substrate-budget delta NEGATIVE: each SKILL.md byte reduction > atlas byte addition (Map relocates lines to Atlas; small overhead for heading + cross-reference but should net-reduce total loaded-byte cost since Atlas is conditionally-loaded vs Map always-loaded)
• AC5: No behavior change — all skill invocations still read the atlas via view_file; runtime directives now codified in atlas instead of Map
• AC6: Substrate-Mutation Pre-Flight (§1.1 in PR body) — /turn-memory-pre-flight decision-tree applied: Step 1 (Universal rule? No — skill-specific runtime declarations) → Step 2 (Skill lifecycle event? Yes — skill-loaded content placement decision) → Atlas placement substrate-correct

Out of Scope

• Restructuring atlas internals (this ticket is Map→Atlas RELOCATION, not Atlas refactor)
• Adding new substrate to any of the 3 skills (cleanup-only)
• Touching SKILL.md routers within floor (23 other skills clean)
• Discussion #11265 graduation work (separate; Issue #11267 is that ticket)
• Behavior changes to the runtime directives themselves (preserve verbatim; relocate only)

Avoided Traps

• Trap: defer to Gemini per AC-CycleE → rejected per operator-correction 2026-05-12 ~13:03Z; Gemini is over-loaded (8/4/1 distribution); rebalancing work AWAY from her is the fair-distribution intent. AC-CycleE non-origin-author is satisfied by ANY non-Discussion-author (any non-Opus); GPT cross-family review fulfills the recursive substrate-validation discipline.
• Trap: split into 3 separate tickets/PRs → rejected; the 3 cleanups are mechanically uniform Map→Atlas relocations; single bounded PR is leaner per substrate-budget AC. Three PRs would worsen the in-flight stack during operator-acknowledged substrate-corruption resolution.
• Trap: opportunistic atlas restructure → rejected; cleanup scope is Map→Atlas relocation only. Restructuring atlas internals is separate future work if needed.
• Trap: add new ACs/guardrails to the affected skills → rejected; preserve substantive content verbatim. Discipline is shorten, not improve.

Related

• Discussion #11265 (graduated to Issue #11267; Cycle 1.7 substrate-corrected) — substrate-source for Layer 1 (skill-substrate-health) durable target; this ticket actions one concrete instance of the Layer 1 work
• Issue #11267 (CLOSED-or-OPEN per Gemini decision post-halt-lift) — graduation ticket primarily for Layer 2 (cross-family rotation discipline); separate from this Layer 1 cleanup work
• /create-skill (skill-authoring-guide.md §"Byte Budget for SKILL.md Routers") — authoritative source for 12-line empirical floor + 7-line lower bound + Map/Atlas placement discipline
• Operator feedback 2026-05-12 ~12:58Z (relayed via @neo-gemini-3-1-pro MESSAGE:40b73fe1) — substrate-direction anchor: "push for shorter and clearer skills. map versus world atlas. query raw memories extensively. this is nothing new."
• Operator correction 2026-05-12 ~13:03Z — fair-distribution anchor: Opus + GPT take cleanup work; do NOT defer to over-loaded Gemini

Cross-Family Review Routing

Primary reviewer: @neo-gpt (cross-family + Gemini halted + AC-CycleE recursive-substrate-validation requires non-origin author OR cross-family implementation-ownership review).

Origin Session ID

c2912891-b459-4a03-b2af-154d5e264df1 (Claude Opus 4.7 / Claude Code 1M context, 2026-05-12 session; operator-direct correction 13:03Z + audit findings 13:00Z + Discussion #11265 Layer 1 substrate-source)

Handoff Retrieval Hints

• query_raw_memories(query="skill substrate health Map Atlas 12-line floor session-sunset lead-role peer-role")
• ask_knowledge_base(query="Map vs World Atlas SKILL.md router byte budget")
• File anchors: .agents/skills/session-sunset/SKILL.md + .agents/skills/lead-role/SKILL.md + .agents/skills/peer-role/SKILL.md + respective references/*.md atlas files + .agents/skills/create-skill/references/skill-authoring-guide.md
• Substrate-source anchor: Discussion #11265 Layer 1 + operator feedback 12:58Z + operator correction 13:03Z

🤖 Generated with Claude Code