Slot rule + disposition taxonomy + byte budget retrofit into create-skill authoring guide

Context

Sub 1 of Epic #10757 — cognitive-load audit cycle 2. No dependencies. This is the creation-time gate (V1) — most contained sub, ideal first delivery.

/create-skill already encodes Map+Atlas at structural level (Progressive Disclosure pattern in skill-authoring-guide.md, 91 lines). What it lacks: the discriminator taxonomy that cycle-1 surfaced for AGENTS.md compaction (now resident in AGENTS.md as the Compaction Taxonomy table).

The Problem

When a new skill is authored today, the author has no taxonomy for deciding which content goes where. Map vs Atlas is a binary; the cycle-1 cognitive-load epic surfaced a richer 3-axis discriminator (trigger-frequency × failure-severity × enforceability) and a 5-disposition taxonomy (keep, move, compress-to-trigger, rewrite, retire). Without this in /create-skill, every new skill starts without the framing that cycle-1 retrofitted into AGENTS.md, forcing the same discriminator to be re-derived per-skill via post-hoc compaction (which is exactly what cycle-1 had to do).

The Architectural Reality

• .agents/skills/create-skill/SKILL.md (7 lines) — router pointing to authoring guide
• .agents/skills/create-skill/references/skill-authoring-guide.md (91 lines) — current authoring substrate
• AGENTS.md Compaction Taxonomy section — codifies the 3-axis slot rule + disposition taxonomy + substrate-vs-discipline tagging (MACHINE-ENFORCEABLE-CANDIDATE / DISCIPLINE-ONLY)
• learn/agentos/measurements/cognitive-load-baseline-2026-05.md — empirical anchor for byte-budget guidance (SKILL.md routers empirically 7-12 lines)

The Fix

Retrofit four additions into skill-authoring-guide.md:

1. 3-axis slot rule as a guidance section before §1 Writing the Router. New skill authors apply the rule to each section they're considering: trigger-frequency (always-loaded vs edge-case-triggered) × failure-severity (catastrophic vs minor) × enforceability (machine-enforceable vs discipline-only).

2. Disposition taxonomy (keep, move, compress-to-trigger, rewrite, retire) enumerated with definitions. Even at creation time, knowing the disposition vocabulary forces the author to declare why each section earns its slot.

3. Substrate-vs-discipline tagging (MACHINE-ENFORCEABLE-CANDIDATE / DISCIPLINE-ONLY) recommended for any section likely to be cited from per-turn substrate. New skills inherit this tagging discipline rather than acquiring it via post-hoc compaction. Pointer to AGENTS.md Compaction Taxonomy as the canonical worked example.

4. Byte budget guidance for SKILL.md routers — empirical floor 7-12 lines (anchor: cognitive-load-baseline-2026-05.md). Authors writing routers exceeding this budget should justify or split.

These are guidance additions, not mechanical gates. The mechanical gate is V2 (mutation-time Pre-Flight) — V1 makes the discriminator visible at authoring time so V2 has substrate to reference.

Avoided Traps

• Mechanical gate at creation time: would block legitimate large authoring guides (this very file). The 4 retrofits are guidance + frame-naming, not gating thresholds.
• Replicating the AGENTS.md Compaction Taxonomy table verbatim: the per-section disposition table is useful at compaction time; at creation time the disposition is implicit (keep for everything authored). The taxonomy table belongs in V2 mutation-time substrate, not V1.
• Lifting the 3-axis rule and the worked example as one block: the rule is universal; specific worked-example dispositions vary per skill. Lift the discriminator framing, not the AGENTS.md content.
• Bloating SKILL.md (the router) instead of the authoring guide: all 4 retrofits go into references/skill-authoring-guide.md; SKILL.md remains 7 lines.

Acceptance Criteria

• (AC1) skill-authoring-guide.md includes a "3-axis slot rule" guidance section before §1 Writing the Router; section explains trigger-frequency × failure-severity × enforceability with one worked example
• (AC2) Disposition taxonomy (keep, move, compress-to-trigger, rewrite, retire) enumerated with definitions
• (AC3) Substrate-vs-discipline tagging (MACHINE-ENFORCEABLE-CANDIDATE / DISCIPLINE-ONLY) recommended; pointer to AGENTS.md Compaction Taxonomy as canonical worked example
• (AC4) Byte budget guidance for SKILL.md routers cited from cognitive-load-baseline-2026-05.md; not a hard cap, a discriminator
• (AC5) Existing /create-skill content preserved; retrofit is additive — Progressive Disclosure framing, folder-structure scaffold, lesson-promotion path, Claude symlink mandate, and verification checklist all retained
• (AC6) Per feedback_no_harness_private_load_bearing_citations discipline (codified this session): no harness-private memory citations in the retrofit text; all references resolve to repo-resident substrate or public anchors

Out of Scope

• V2 mutation-time gate (/pull-request Pre-Flight) — separate sub
• V3 cron audit primitive — separate sub
• V4.* MCP tool surface — separate sub-cluster (V4.1 supersedes #9953)
• Refactoring the existing skill-authoring-guide.md Sections 1-4 — additive only
• Mechanical gates / linting against the slot rule — V2 territory

Related

• Parent epic: #10757
• Empirical baseline: learn/agentos/measurements/cognitive-load-baseline-2026-05.md
• AGENTS.md Compaction Taxonomy section — canonical reference for slot rule + disposition + tagging
• Sibling subs (TBD as filed): V2a, V2b, V3, V4.1-V4.4
• Sibling tickets: #10756 (V5), #9953 (subsumed by V4.1)
• Predecessor cycle-1 epic: #10733 (closed RECOMMEND_CLOSE_COMPLETED)

Origin Session ID: 23b9cbcd-4938-4a46-b21a-0d48dd12e7e7

Retrieval Hint: query_raw_memories(query="cognitive-load cycle-2 V1 slot rule retrofit create-skill skill-authoring-guide disposition taxonomy byte budget 10757")