Repository navigation
perf(discovery): cut the research skill body through progressive disclosure #12659
Description
Refs #12243
Blocked by #12647
Outcome
The /discovery:research body is cut through progressive disclosure: action-specific detail moves to reference files read on demand, and the hub keeps routing and gates. This is one of five slices under #12243; the first cut (/source-control:worktree) landed in #12387, whose recipe this follows.
Size at origin/main 3bd42cabfe: body 44,641 bytes (SKILL.md with frontmatter stripped, the method #12387 used); the skill directory holds 302,817 tracked bytes. The binding limit is the paired eval run each cut needs, so each skill is its own PR.
Sequencing: PR #12647 (for #12631) edits this SKILL.md now, and a contract test pins a sentence inside a 20,000-byte window of it. Start after #12647 merges, and keep or repoint that window test as part of the cut.
Done when
- The hub's body bytes fall, measured the same way. Moved text sits verbatim in
reference/, and every citation of it (tests, contract windows, hooks, scripts, fixtures, README) is repointed. -
/docs-hygiene:compress comparereturns VERDICT PASS with no SEMANTIC LOSS. - The skill's cases run on the before and after copies with
/evals:plugin-eval --baseline <before.json> --margin <m>, on a roundrun-validity.pymarks VALID on both arms, and the after arm is non-inferior (plugins/playbooks/skills/skill-authoring/reference/skill-criteria.md, "Cutting text"). Only one case inplugins/discovery/evals/targets this skill (rs-deep-offers-native-workflow-first); add cases that read the moved text before the before-run.
Related
- perf: cut the bodies of the most-invoked skills through progressive disclosure - part 1 of 2 #12243 (parent), perf(source-control): move the worktree nesting measurement out of the skill hub #12387 (cut 1 recipe), discovery research SKILL.md: a contract-tested sentence sits 1 byte inside its 20000-byte window #12631, fix(discovery): give the research Effort-ceiling sentence headroom in the re-attach slice #12647, perf: cut the bodies of the most-invoked skills through progressive disclosure - part 2 of 2 #12244 (after figure).
Activity
github-actions commentedon Nov 3, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Medium
Complexity: Medium
Uncertainty: Knowable
Flags: none
Blocked by: none
Understood: cut the
/discovery:researchSKILL.md body by moving action-specific detail verbatim intoreference/files. The hub keeps routing and gates. Every citation of the moved text gets repointed, following the #12387 recipe. Done-when is a smaller body, a/docs-hygiene:compress comparePASS, and a non-inferior paired eval. The eval step needs a host run, so it is routed separately.Notes: #12647 is named as the sequencing dependency. It isn't in the open-issue list, so I listed no blocker. Please confirm it has merged before work starts. No duplicates found.
Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Nov 3, 2026
Not admitted to the issue lane: its done-when asks for what an unattended lane run cannot perform: item 3 (a model eval run: "The skill's cases run on the before and after copies with /evals:plugin-eval ..."); recommended split: keep the other items here as the admissible issue and move these to a child issue for the host queue (
host-run); it carries host-run. Admission takes the issue once each of those is met, when intake triages it again (.github/actions/resolve-config/admission-rules.json).melodic-automation-lanes commentedon Nov 3, 2026
The latest triage verdict holds nothing: admission moved the issue to
status: confirmed; its comment says why.kyle-sexton commentedon Nov 3, 2026
Claim. Backlog HITL host-run worker is taking this now.
kyle-sexton commentedon Nov 3, 2026
Negative control, preregistered before the first paid run (
rewrite-a-plugin.mdstep 4).skills/research/reference/practitioner-test.md; the hub keeps a pointer.rs-*cases:rs-deep-offers-native-workflow-first(existing) and two new ones that read the moved text,rs-practitioner-needs-shipped-workandrs-profile-imperative-is-red-flag. All graders are deterministic (regex,tool_used); no case grants Bash.skills/research) with the researchSKILL.mdbody deleted and its frontmatter kept;context/andreference/kept.claude-opus-5-5,--case 'rs-*' --runs 2 -j 2 --ablation none --keep-temp, launched together with a before run.--baseline before.json) calls acase drop, or the mean case score falls by 0.15 or more. Neither means the body is unmeasurable by this suite, and the before text ships.run-validity.py --own-copy-search-warnson the degraded side only.noise-report.py --baseline before.json --margin 0.05, VALID on both arms.docs(discovery): repoint the aiforui L9 row to the practitioner test …