refactor(skills): slim over-budget SKILL.md bodies via progressive disclosure - #117
Merged
Conversation
…sure
WL3-810 P5 — baseline-4 skill budget ("skills <10KB each").
python-testing SKILL.md was 52.6KB / 916 lines: 36 standalone pitfall
write-ups (with WRONG/RIGHT code and dated evidence) all inline in the
always-loaded SKILL.md. Split into an index + references, nothing deleted:
- SKILL.md: 52.6KB -> 5.2KB. Keeps When to Use + 3 global disciplines +
a 36-row symptom->cause->ref index + verification checklist.
- references/python-testing-pitfalls.md: all 36 pitfalls (+ 4 sub-entries)
preserved verbatim under §1..§36 anchors the index points at.
- config/skill-provenance.yaml: source_sha256 recomputed (CRLF-normalized),
version 1.0.0 -> 1.1.0.
Verified: SKILL_PROVENANCE_PASS skills=13; measure_p5 shows the SKILL.md
back under 10KB.
…t-environment
WL3-810 P5 — complete the baseline-4 skill budget ("skills <10KB each") for
the remaining two over-budget managed skills.
agent-workflow-fortress (21.9KB -> 5.7KB SKILL.md):
- Keeps Overview / When to Use / Core Loop / Safety Rules / Stack Boundary /
Real-task loop rule / Verification Checklist + a reference index.
- Moved to references/ (verbatim): absorption-rules.md (8 absorption rules),
context-token-hygiene.md (7-step cleanup), skill-library-overlap-audits.md
(8-step audit), autonomy-protocols.md (13-step iteration protocol + parallel
autonomy + handoff + task tickets + deterministic slices).
The existing 5 absorption references are now indexed from SKILL.md.
windows-development-environment (31.9KB -> 8.6KB SKILL.md):
- Keeps the core selection/quoting/resolution/server/deployment strategies +
Common pitfalls + Verification.
- Moved the 17 dated "(validated ...)" case studies to
references/validated-cases.md (verbatim command recipes).
config/skill-provenance.yaml: source_sha256 recomputed (CRLF-normalized) for
both, versions 1.4.7->1.5.0 and 1.2.1->1.3.0.
Verified: SKILL_PROVENANCE_PASS skills=13; measure_p5 shows all 13 SKILL.md
bodies under 10KB (skill_md_over_10kb=[]).
…pment-environment The slim rewrite dropped two strings that test_workflow_governance.py locks as safety contracts: - "### PowerShell selection policy" heading (test_windows_skill_does_not_bypass_provider_or_credential_boundaries) - "Do not assume `python` and `python3` resolve to the same interpreter." + "Hermes workflow scripts use `python`" (test_windows_skill_requires_explicit_interpreter_selection) Restore both verbatim; recompute source_sha256 (54ad5d6e) and regenerate CURRENT_STATE (digest a2be3d27). Verified: governance gate 795 tests OK, SKILL_PROVENANCE_PASS skills=13.
…r fields
Two drift fixes to the single config-ownership authority (WL3-200 / WL3-330):
1. layers.USER_OVERLAY.behavior said "14 Skills"; the authoritative count is
13 (test_skill_package_digest.test_managed_set_is_exactly_thirteen,
managed-config-schema.yaml owned_asset_roots). Correct to 13.
2. Five-dimension baseline 5 ("full-power model") had no field-level ownership.
Register four OBSERVE fields so the "no downgrade" baseline is auditable:
- hermes.agent.reasoning_effort (official default medium or higher)
- hermes.model.temperature (no downgrade below official baseline)
- cc_switch.provider.cost_multiplier (official 1.0, no cap)
- cc_switch.provider.rate_limits (no daily/monthly caps)
Verified: test_config_ownership 10/10, CONFIG_OWNERSHIP_PASS fields=55.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
WL3-810 P5 + WL3-200/330 consolidation — three "slim/consolidate" changes in one
coherent close-out:
1. Slim over-budget managed skills (WL3-810 P5, baseline-4 budget)
Pure "move, don't delete": every pitfall / case study / rule is preserved
verbatim in
references/, the SKILL.md body keeps only the always-neededtrigger + core loop + safety rules + a symptom→ref index.
references/python-testing-pitfalls.md(§1–§36 + 4 sub-entries)references/validated-cases.md(17 dated case studies)Contract-locked safety markers restored verbatim (PowerShell selection policy,
explicit interpreter selection) — governance gate 795 tests pass.
2. Global config consolidation (WL3-200 / WL3-330)
config-ownership.jsonlayers.USER_OVERLAY.behavior: "14 Skills" → "13Skills" (authoritative count is 13 per
test_managed_set_is_exactly_thirteen).hermes.agent.reasoning_effort,hermes.model.temperature,cc_switch.provider.cost_multiplier,cc_switch.provider.rate_limits.Why
always-needed instructions stay in SKILL.md; bulky/branch-specific detail is
read on demand from
references/.field-level ownership instead of living only in prose.
Verification
SKILL_PROVENANCE_PASS skills=13(source hashes CRLF-normalized).measure_p5: all 13 SKILL.md bodies <10KB —skill_md_over_10kb=[].run_quality_gate.py governance: 795 tests OK (skipped=8).test_config_ownership: 10/10;CONFIG_OWNERSHIP_PASS fields=55.generate_current_state.py --check-current: FRESHNESS_PASS.