LearnNewsExamplesServices
Frontmatter
titledocs(agentos): compress ideation sandbox workflow (#13540)
authorneo-gpt
stateClosed
createdAtJun 19, 2026, 3:09 PM
updatedAtJun 19, 2026, 3:13 PM
closedAtJun 19, 2026, 3:12 PM
mergedAt
branchesdevcodex/13540-compress-ideation-sandbox
urlhttps://github.com/neomjs/neo/pull/13555
Closed
neo-gpt
neo-gpt commented on Jun 19, 2026, 3:09 PM

Resolves #13540 Related: #10757

Compresses the ideation-sandbox workflow from a 25 KB hot payload into a shorter map while preserving the Discussion-first path, OQ tags, Double Diamond gate, Step-Back sweep, Signal Ledger quorum, dissent/liveness sections, and PR-merge-gate handoff. This is compression only: no new policy, no new skill file, and .agents/skills/ideation-sandbox/SKILL.md stays unchanged.

Evidence: L1 (static skill-substrate diff, byte-count audit, grep-visible contract anchors, manifest lint) -> L1 required (doc-only compression ACs). No residuals.

Deltas from ticket

No scope additions. The implementation chose in-place rewrite instead of sibling extraction because a split would risk relocation-only bloat; the normal ideation-sandbox invocation now loads less text and the skill-local aggregate is net-negative.

Load-Effect Audit

Slot rationale: modified section disposition remains compress-to-trigger, but the hot workflow is now a compact map with audit pointers for detailed rationale. The load/placement changed because the previous file mixed routine gates with empirical rationale and repeated consensus details.

  • .agents/skills/ideation-sandbox/references/ideation-sandbox-workflow.md: 25,000 -> 11,941 bytes.
  • .agents/skills/ideation-sandbox/**/*.md: 25,729 -> 12,670 bytes.
  • .agents/skills/**/*.md: 587,207 -> 574,148 bytes.
  • Markdown delta: one existing file, 100 insertions / 146 deletions.
  • New skill Markdown files: none.

Preserved anchors include create_discussion, add_message, [RESOLVED_TO_AC], [GRADUATED_TO_TICKET], Double Diamond, STEP_BACK, all eight Step-Back sweep names, [GRADUATION_APPROVED], [GRADUATION_DEFERRED], [GRADUATION_ABSTAIN], [AUTHOR_SIGNAL], ## Signal Ledger, ## Unresolved Dissent, ## Unresolved Liveness, ## Discussion Criteria Mapping, revalidationTrigger, and pull-request-workflow.md §6.1.1.

Test Evidence

  • git diff --check
  • node ./ai/scripts/lint/lint-skill-manifest.mjs --base origin/dev
  • node ./ai/scripts/lint/lint-skill-manifest.mjs --report-sizes --top 8
  • rg anchor sweep over the edited workflow for the preserved Discussion/graduation/quorum contract terms.

Post-Merge Validation

  • Next ideation-sandbox invocation still routes to the required Discussion/graduation/quorum gates with the lower hot-path payload.
  • Skill size report keeps ideation-sandbox-workflow.md below the top hot-payload ranks after merge.

Commits

  • 85ba7cc89 — compress ideation-sandbox workflow payload.

Authored by Euclid (GPT-5, Codex Desktop). Session 019ed42c-f8fc-7e01-a1a1-a8b5bbf58b64.

neo-gpt
neo-gpt commented on Jun 19, 2026, 3:11 PM

Withdrawing this PR.

Operator review caught the core issue: this over-compressed the ideation-sandbox workflow and removed meaning from a flagship skill. The result is not semantically equivalent to the original, so it should not proceed to review or merge.

I am closing the PR rather than iterating on this diff. A future attempt at #13540 needs a much narrower preservation-first strategy: measure load, but protect the full design intent before deleting prose.