Codify decision-atom lesson promotion

Context

Parent epic #11605 defines the skill-baggage compression program. Its first two acceptance criteria require codifying the Bias -> Rule -> Rationale -> Trigger atom and updating /create-skill so future lesson promotion does not default to blind payload append.

This child exists to make that first slice claimable without dragging the inventory and pilot migrations into the same PR.

Duplicate / adjacency sweep before filing:

• Repo-local grep for Bias -> Rule -> Rationale -> Trigger, Lesson Promotion Path, decision-atom, and decision atom found #11605, the existing /create-skill Lesson Promotion Path, and unrelated historical references; no dedicated child issue for this AC1/AC2 slice.
• Live GitHub search for "Bias -> Rule -> Rationale -> Trigger" OR "Lesson Promotion Path" OR "decision atom" returned #11605 and broad cloud epic #9999 only.
• ask_knowledge_base() semantic sweep was attempted and failed with Internal error: Error finding id; this ticket records that limitation instead of treating the failed KB call as evidence.

The Problem

/create-skill currently tells agents to promote valuable lessons by updating the appropriate skill payload or authoring a new skill. That is directionally correct, but too coarse for the failure class captured in #11605: lessons that contain historical incident baggage, provenance, or rare edge-case stories are likely to be appended into runtime workflow manuals instead of compressed into a small operating atom with a trigger pointer to deeper atlas/provenance material.

The missing contract is not a new migration of every skill. It is the authoring rule for future lesson promotion:

Bias: <default wrong model>. Rule: <correct action>. Rationale: <why this matters>. Trigger: <when to load deeper atlas, if needed>.

Without that contract, each future friction-to-gold cycle can keep accreting long prose into high-frequency skill payloads.

The Architectural Reality

Current source-of-authority verification:

• learn/agentos/decisions/0008-skill-anatomy-and-authoring-contract.md defines skills as Progressive Disclosure substrate, with SKILL.md as a lightweight Map and references/*.md as Atlas payload.
• ADR 0008 §2.2 already says Map vs World Atlas applies recursively to workflow files when they exceed natural load-frequency boundaries.
• .agents/skills/create-skill/references/skill-authoring-guide.md currently contains the Lesson Promotion Path, but it still says to update the payload with the new lesson/trap/rule and does not require the decision atom or atlas/provenance routing decision.
• learn/agentos/decisions/0007-agents-md-compaction-taxonomy.md is the broader compaction-taxonomy authority for modifications to loaded instruction substrate.

Therefore this work should amend ADR 0008 or add a narrow successor/companion decision only if ADR amendment is too crowded, then update the /create-skill payload as the procedural HOW-TO. It should not add another always-loaded AGENTS.md rule.

The Fix

1. Amend ADR 0008, or create the smallest successor/companion ADR if amendment would obscure the existing skill-shape decision, to define Bias -> Rule -> Rationale -> Trigger as the preferred runtime-instruction atom for promoting operational lessons into skill substrate.
2. Update .agents/skills/create-skill/references/skill-authoring-guide.md Lesson Promotion Path so new lessons route through:
   • runtime atom when the behavior should fire during normal skill use;
   • atlas/provenance payload when details are history, examples, or rare edge cases;
   • new skill only when the domain is genuinely new.
3. Keep /create-skill/SKILL.md router unchanged unless the existing trigger wording is empirically insufficient.
4. Preserve behavior: the new guidance must still make valuable lessons durable, while preventing payload append as the default shape.

Contract Ledger Matrix

Decision Record Impact

amends ADR 0008 unless implementation proves a narrow successor/companion ADR is cleaner. Aligned with ADR 0007 compaction taxonomy.

Acceptance Criteria

• ADR 0008 or a narrow successor/companion ADR defines Bias -> Rule -> Rationale -> Trigger and when to use it.
• The decision text distinguishes runtime instruction atoms from atlas/provenance material.
• /create-skill Lesson Promotion Path routes new lessons through the atom/atlas/new-skill decision instead of defaulting to payload append.
• /create-skill/SKILL.md remains unchanged, or any router wording change is minimal and justified by trigger clarity.
• No broad migration of pr-review, pull-request, or other skills is included in this child.
• Validation includes static grep for the new atom, git diff --check, and any relevant skill/manifest lint if touched.

Out of Scope

• Inventorying all existing skills for baggage.
• Migrating pr-review, pull-request, or any other hot-path skill.
• Creating incident/provenance directory structure beyond a pointer needed to explain the routing decision.
• Adding AGENTS.md rules.
• Closing parent #11605.

Avoided Traps

Related

• Parent epic: #11605.
• Adjacent: #11604, #11603, #11319, #10537.
• Source authorities: ADR 0007, ADR 0008, /create-skill skill payload.

Origin Session ID: 7fa9a0db-5c77-4584-b7a5-7693c7ef90c0

Handoff Retrieval Hint: query_raw_memories("#11605 Bias Rule Rationale Trigger create-skill Lesson Promotion Path ADR 0008 skill baggage compression")