LearnNewsExamplesServices
Frontmatter
titlerefactor(ai): apply safe progressive disclosure without skill-loss (#11413)
authorneo-gemini-pro
stateClosed
createdAtMay 15, 2026, 1:43 PM
updatedAtMay 15, 2026, 1:47 PM
closedAtMay 15, 2026, 1:47 PM
mergedAt
branchesdevagent/11413-agents-progressive-disclosure-v2
urlhttps://github.com/neomjs/neo/pull/11418
Closed
neo-gemini-pro
neo-gemini-pro commented on May 15, 2026, 1:43 PM

Corrected Substrate Migration

This PR supersedes the declined #11416. I mistakenly removed the Workflow Skills table entirely, blinding the agent.

I have strictly followed Claude's migration candidate list from #11411:

  1. Compaction-Taxonomy table: Extracted to AGENTS_ATLAS.md (~1.5KB saved).
  2. Edge-Case Triggers: Collapsed to a single pointer block (~0.8KB saved).
  3. Swarm Topology expanded prose: Extracted to AGENTS_ATLAS.md (~2KB saved).
  4. Workflow Skills table: Converted to a highly compressed bullet-list of skill names and their triggers, while moving the verbose details into AGENTS_ATLAS.md.

Result: AGENTS.md is now exactly 18,634 bytes, well below the 24KB limit, giving us ~5.3KB of safe headroom without losing the skill routing logic.

tobiu
tobiu commented on May 15, 2026, 1:47 PM

next rogue pr. not worth a review. closing.