Compress Double Diamond skill rules into Atlas payloads

Context

PR #11095 merged the Double Diamond graduation guard from #11086. That guard is valuable substrate, but it added dense process-rule bodies directly to three frequently-read workflow maps:

• .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md §5.1
• .agents/skills/ticket-create/references/ticket-create-workflow.md §1c
• .agents/skills/epic-review/references/epic-review-workflow.md Stage 2

PR #11097 just proved the companion pattern on #11085 §9.0: keep the trigger and decision rule in the map, move case-study depth / trigger catalogs / bias rationale to granular Atlas payloads, and report byte deltas. This ticket applies that same discipline to the now-merged #11095 content.

The Problem

The Double Diamond guard deliberately adds process friction to prevent premature convergence. If the whole case study and edge-case rationale stays inline in high-level workflow maps, we reintroduce the skim-risk that Progressive Disclosure is meant to prevent.

This is not an argument to remove the guard. The rule came from real substrate failure (#11076 -> #11077, #11082 / PR #11083, #11078), and it should remain easy to notice. The issue is placement: routine skill readers need the trigger, required action, and pointer; only agents who hit the edge case need the full archaeology.

The Architectural Reality

• .agents/skills/create-skill/references/skill-authoring-guide.md defines the Map vs World Atlas placement constraint and 3-axis slot discriminator.
• PR #11097 is the direct precedent: pr-review-guide.md kept a short §9.0 trigger while pr-review/audits/cycle-1-premise-preflight.md became the Atlas payload.
• Current post-merge wc -c sizes on origin/dev after PR #11095/#11097/#11096 merges:
  • ideation-sandbox-workflow.md: 9,448 bytes
  • ticket-create-workflow.md: 15,285 bytes
  • epic-review-workflow.md: 17,405 bytes
• The new content is docs/skill substrate only; no runtime code or tests should be involved.

The Fix

Run a focused compression pass over the #11095 additions:

1. Classify each new Double Diamond section with the create-skill 3-axis discriminator and disposition taxonomy.
2. Keep concise trigger / decision-rule / cross-link text in each workflow map.
3. Move long rationale, empirical anchors, examples, and edge-case mechanics into one or more granular Atlas payload files.
4. Preserve behavior: agents must still know when the Double Diamond gate fires, what blocks graduation/ticket creation/epic review, and where to read the full rationale.
5. Measure wc -c / blob-size before and after for every touched map and Atlas file.

Contract Ledger Matrix

Acceptance Criteria

• The #11095 Double Diamond additions are audited and each touched section receives a keep, compress-to-trigger, move, rewrite, or retire disposition.
• Long empirical anchors / examples / bias or source-authority rationale are moved to Atlas payloads when doing so reduces loaded map bytes without hiding the trigger.
• Each remaining workflow-map entry is short enough to read in one pass and includes a direct relative link to the Atlas payload when extraction occurs.
• Behavior is preserved for all three surfaces: ideation graduation, ticket creation, and epic review.
• No top-level SKILL.md router gains heavy content.
• PR body reports before/after byte counts for every touched map and Atlas file.
• If any #11095 section is intentionally left inline, the PR body explains why extraction would reduce clarity or behavior.

Out of Scope

• Changing Double Diamond semantics from #11086 / PR #11095.
• Re-opening #11090 / PR #11097.
• Broad cognitive-load cleanup of all skill payloads (#10733 owns that).
• Runtime code, daemon code, tests, or issue-sync behavior.

Avoided Traps / Gold Standards Rejected

• Delete or weaken the guard because it is long: rejected. The process failure was real; compression must preserve behavior.
• Move every sentence automatically: rejected. The map must still carry enough trigger surface to fire reliably.
• Split without measuring: rejected. The PR must state before/after byte deltas, including any Atlas payload growth.
• Bundle with #11092 or #11094: rejected. This is a small GPT-sized follow-up and should not collide with Opus/Gemini lanes.

Related

• PR #11095 / issue #11086 — Double Diamond guard implementation.
• PR #11097 / issue #11090 — precedent compression pass for #11085 §9.0.
• Discussion #11079 — source Discussion for the Double Diamond guard.
• Discussion #11091 — public source for operator-as-peer / context-asymmetry refinement cited in §1c.
• Epic #10733 — broad cognitive-load cleanup; this ticket remains narrow.

Origin Session ID: 20a824b0-29d1-4082-ae12-87705ec69c3f

Retrieval Hint: "post-11095 Double Diamond Map vs Atlas compression ideation-sandbox ticket-create epic-review"