Compress ideation-sandbox workflow payload

Context

Release classification: post-release hardening / Agent OS cognitive-load reduction; boardless.

This ticket continues the skill-bloat friction-to-gold chain under #10757.

Live evidence from the current reduction pass:

• PR #13530 was closed unmerged after it attempted another additive skill-work shape (+38/-1 across three skill files).
• PR #13534 / #13533 adds the mechanical guard for net-positive .agents/skills/**/*.md growth.
• PR #13536 / #13535 starts the inverse move by reducing post-review-pickup instead of adding more prose.
• #13537 and #13538 queue the two largest remaining hot workflow reductions.
• node ./ai/scripts/lint/lint-skill-manifest.mjs --report-sizes --top 20 reports aggregate skill Markdown at 614586 bytes / 7319 lines after the first reduction branch.
• The same report ranks .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md as 25000 bytes, 201 lines, 152 signals, disposition compress-to-trigger.

Live latest-open sweep: checked the latest 20 open GitHub issues on 2026-06-19; no equivalent ideation-sandbox workflow compression ticket was found. The nearest siblings are #13537 (pr-review) and #13538 (pull-request). A2A in-flight sweep: latest 30 messages showed no competing [lane-claim] / [lane-intent] for this exact scope. Local exact sweep found only the existing audit pointer .agents/skills/ideation-sandbox/audits/double-diamond-divergence-guard.md and historical discussion anchors, not an active issue.

Progressive Disclosure check: consulted .agents/skills/create-skill/references/skill-authoring-guide.md; this ticket is intentionally about net reduction, not relocating the same text behind a new file.

The Problem

The ideation-sandbox skill is one of the remaining large governance payloads. It matters because high-blast proposals already require heavy peer-review and graduation discipline; making that workflow heavier increases the odds that smaller-context agents either skip the payload, compact mid-lifecycle, or follow only the nearest recent paragraph.

The current file sits exactly at the default 25 KB payload budget and carries a high signal count. That is a smell for a workflow map: it likely mixes always-relevant graduation mechanics with historical rationale, repeated guard language, and edge-case detail that should either be compressed or conditionally triggered.

This is the same failure mode that closed #13530 unmerged: a real workflow problem exists, but the wrong fix is another paragraph in a hot skill. The right fix is to preserve the behavior while reducing the routine load.

The Architectural Reality

The target surface is .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md.

The skill authoring guide says workflow files are maps too: always-relevant sections stay inline; edge cases become one-line triggers only when extraction truly reduces invoked load. It also says tool mechanics belong in tool descriptions, not skill payloads, and bloat must be measured against the smallest-context peer.

Discussion #11598 already names the recurring trap: symptom-by-symptom fixes create more tickets, more PR churn, and more skill bloat without closing the root cause. This ticket applies the opposite direction to the ideation workflow itself.

The Fix

Compress .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md while preserving the ideation behavior contract.

The implementation should:

• Keep .agents/skills/ideation-sandbox/SKILL.md unchanged unless a one-line trigger correction is strictly required.
• Reduce duplicated graduation, quorum, step-back, and discussion-routing prose.
• Preserve the minimum behavior required for high-blast proposals:
  • exploratory ideas start in Discussion, not Issue;
  • unresolved ambiguity stays in the sandbox;
  • graduation requires source-bound acceptance criteria and quorum where applicable;
  • ticket graduation does not bypass later PR review;
  • public ledgers remain source-of-authority for convergence.
• Avoid relocation-only extraction. A new sibling payload is acceptable only when the total .agents/skills/ideation-sandbox/**/*.md byte delta is negative and the PR proves the normal ideation invocation loads less text.
• Avoid adding new governance policy; this is a compression ticket, not a semantics-expansion ticket.

Contract Ledger Matrix

Decision Record impact

Aligned with ADR 0008 (SKILL.md anatomy / Progressive Disclosure) and #10757 (cognitive-load reduction). No ADR amendment expected.

Acceptance Criteria

• Record before/after byte counts for .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md.
• The PR has a net-negative .agents/skills/ideation-sandbox/**/*.md byte delta.
• The PR has a net-negative .agents/skills/**/*.md byte delta.
• No new skill Markdown file is added unless the aggregate skill-local delta remains negative and the PR explains why the split reduces normal invocation load rather than relocating bloat.
• The ideation behavior anchors listed in The Fix remain grep-verifiable.
• node ./ai/scripts/lint/lint-skill-manifest.mjs --base origin/dev passes.
• The PR body includes a load-effect audit for any touched skill substrate.

Out of Scope

• Changing the ideation-sandbox policy itself.
• Adding a new skill.
• Reopening PR #13530.
• Compressing pr-review or pull-request; those are tracked separately in #13537 and #13538.
• Changing the net-positive growth guard in #13533.

Avoided Traps

• Another symptom paragraph: rejected because it is the loop this ticket exists to stop.
• Relocation-only extraction: rejected because smaller-context agents still pay the same aggregate substrate cost when the skill is invoked.
• Weakening graduation discipline to save bytes: rejected; the target is clearer compression, not behavioral retreat.

Related

• Parent: #10757
• Guard: #13533
• First reduction: #13535 / PR #13536
• Sibling reductions: #13537, #13538
• Closed additive precedent: PR #13530

Origin Session ID: 0a1dbe52-d3d0-43c4-8eb5-53a9e8499236

Handoff Retrieval Hints

Search patterns:

• skill bloat ideation-sandbox workflow compress-to-trigger
• ideation-sandbox-workflow 25000 bytes 152 signals
• symptom-by-symptom fix skill bloat Discussion 11598
• smallest-context peer 258k skill payload bloat