Repository navigation
Share execution rules and shorten skills below 12 KB - #369
briansuchy wants to merge 6 commits into
Conversation
egm-triage-bot
left a comment
There was a problem hiding this comment.
Automated first-pass review found zero blocking or major issues. Routing to @google/egm-maintainers for maintainer review.
2ba8584 to
943a259
Compare
egm-triage-bot
left a comment
There was a problem hiding this comment.
Automated first-pass review found zero blocking or major issues. Routing to @google/egm-maintainers for maintainer review.
egm-triage-bot
left a comment
There was a problem hiding this comment.
Automated first-pass review found zero blocking or major issues. Routing to @google/egm-maintainers for maintainer review.
n3wth
left a comment
There was a problem hiding this comment.
Reviewed 0928e5b.
[P2] Please retain minimal inline constraints in bootstrap, resolve, review-loop and drift-report when file tools are unavailable. These skills now say “proceed inline” after moving their shared rules to resources/execution_instructions.md, so that fallback cannot enforce rules it never reads. Brainstorm, draft and review already keep an inline fallback; please use the same approach in these four and their packaged copies. Please rebase before the next review.
…ion, and restore audit invariants - Anchor resources/execution_instructions.md and docs/harnesses.md as relative Markdown links across all 7 skills (with CLAUDE_PLUGIN_ROOT, ~/.gemini pathless <SKILL> fallback, and toolless inline fallback), and enforce anchored links in test_harnesses.py (F1). - Require interactive question gates to emit non-empty chat text first and run as the sole tool call in their model response step after earlier preparatory file writes/reads finish (F2, F5C). - Exempt egm-resolve Step 4 and Interactive breakdown zero-tool plain-chat prompts from structured question tool invocation, and clarify egm-review Stages 7 and 9 checkpoint-first routing across all non-uninterrupted modes (F3A, F3B). - Wire resources/execution_instructions.md into all 7 skills/*/test_scenarios.yaml rule lists, retain compact inline banned-word tuples, and assert rule file presence in test_scenario_coverage.py (F4). - Restore egm-bootstrap template names, chunking/JSON schema details, and contract drift verification targets; egm-brainstorm 7-lens persona paths and RFC feedback pause; and egm-draft cumulative section append wording while keeping all 7 skill bodies under 12 KB (F5A, F5B, F5C). - Regenerate distributions/claude/ via scripts/pack_plugin.py --target claude.
…ignore, and draft/scratchpad guards
…, review-loop, and drift-report
0928e5b to
731796c
Compare
|
Rebased onto |
Summary of Key Changes
Closes #371
On
origin/main(0997679),skills/egm-bootstrap/SKILL.md(23,931body B) andskills/egm-review/SKILL.md(23,887body B) sit within69B and113B of Antigravity's24,000-byte skill body limit (tests/unit/test_skill_load_limits.py::test_skill_body_fits_antigravity), whileskills/egm-draft/SKILL.md(17,115body B) andskills/egm-brainstorm/SKILL.md(14,812body B) carry duplicated boilerplate for delegation, artifact-first delivery, plain-English style, anti-sycophancy word lists, and modal invariants.This PR refactors shared behavioral constraints into
resources/execution_instructions.mdacross all seven skills and compacts all oversizedSKILL.mdfiles in place (without splitting workflows into newreferences/*.mdfiles), bringing totalSKILL.mdbody volume down by-37,480 B(-37.3%) and keeping every skill under12,000 B(at least12,018 Bof headroom under the24,000-byte cap). The packaged Claude runtime underdistributions/claude/is regenerated to match source byte-for-byte (scripts/pack_plugin.py --target claude).distributions/claude/mirror)origin/mainBody / Raw)0928e5bBody)skills/egm-bootstrap/SKILL.md23,931 B/24,090 B11,439 B(-52.2%)-12,492 B69 B -> 12,561 B432 -> 94skills/egm-brainstorm/SKILL.md14,812 B/15,205 B7,863 B(-46.9%)-6,949 B9,188 B -> 16,137 B271 -> 104skills/egm-draft/SKILL.md17,115 B/17,479 B10,379 B(-39.4%)-6,736 B6,885 B -> 13,621 B321 -> 82skills/egm-review/SKILL.md23,887 B/24,257 B11,982 B(-49.8%)-11,905 B113 B -> 12,018 B426 -> 101skills/egm-review-loop/SKILL.md9,846 B/10,325 B9,691 B(-1.6%)-155 B14,154 B -> 14,309 B199 -> 191skills/egm-resolve/SKILL.md8,205 B/8,541 B8,647 B(+5.4%)+442 B15,795 B -> 15,353 B187 -> 188skills/egm-drift-report/SKILL.md2,805 B/3,169 B3,120 B(+11.2%)+315 B21,195 B -> 20,880 B67 -> 68100,601 Bbody63,121 Bbody-37,480 B(-37.3%)<= 11,982 B1,903 -> 828resources/execution_instructions.md2,554 B4,877 B+2,323 B37 -> 721. Pure Deduplication & In-Place Compaction (Behavior-Preserving)
resources/execution_instructions.md: ConsolidatedArtifact-First Delivery,Plain-English Standard,Adversarial Rigor (No Sycophancy)word lists,Subagent Delegation (Goldfish Isolation)and nesting fallbacks, single-selectInteraction Invariants, andVCS IsolationintoShared Execution Constraints (All Skills), linked from all sevenSKILL.mdfiles via anchored relative Markdown links ([execution instructions](../../resources/execution_instructions.md)),${CLAUDE_PLUGIN_ROOT}/resources/execution_instructions.mdfor Claude Code,~/.gemini/config/plugins/egm/resources/execution_instructions.mdfor pathless<SKILL>CLI injection, and toolless inline fallbacks (plussystem_prompt.ruleswiring across all sevenskills/*/test_scenarios.yamlsuites).Design-Only Scratchpad Rulestrictly toegm-draft(preservingtests/unit/test_harnesses.py::test_scratch_files_use_private_directoriessubstrings, active-scratch-dir preference beforemktemp -d, and anti-implication guard) and kept drafting-only pacing rules (Explicit Consensus Gates,ONE QUESTION AT A TIME,NO SUMMARIES,Contradiction Auditing,Author Decides Architecture) insideskills/egm-draft/SKILL.mdso non-drafting skills never inherit conflicting rules.question/optionsspecs, foldedegm-reviewStages5-9into one table (with direct-dispatch standard path mapping and retry-once/Failed Reviewersfallback), and compactedegm-bootstrap(preserving#409's.NET / C#discovery,.razor/.cshtmlextensions,dotnet testguidance, general hidden/system directory exclusion, and swarm template cues) andegm-brainstormwhile preserving extension lists, test exclusions, Bazel target patterns (/...,:all), safety gates,goldfish-explorerschemas and chunking rules, persona paths (agents/<file>.md),concepts_brief.mdtemplates, cumulativedesign_draft.mdsection appends, and standalonetest_scenarios.yamlanchors.distributions/claude/viascripts/pack_plugin.py --target claudeand added regression checks intests/unit/test_harnesses.pyandtests/unit/test_scenario_coverage.pyrejecting unanchoredresources/ordocs/backtick paths and verifyingsystem_prompt.ruleswiring.2. Reconciled Internal Discrepancies & Semantics-Clarifying Refinements
egm-review(Uninterruptedvs. Checkpoint modes & Stage 8 context): UnifiedSubflow: Uninterrupted Full Flowinto theStage Specifications (Stages 5-9)table so document standards and Stage 8'szoom_out_report.mdprior-report context apply consistently in all modes, while clarifying Stages7&9so all non-uninterruptedmodes (interactive,feature_spec,implementation,7,9) writecritic_report.md/implementation_readiness_report.mdand present the 4-option checkpoint gate first (runningegm-resolvewhen Option 1 is selected, then advancing or re-running the stage with max 1 re-audit).egm-review(Step 4: Staging Strategyheading removed): RoutesStep 2/Step 3directly toStage 5..9without changing stage numbers5..9.execution_instructions.md,egm-draft,egm-review,egm-brainstorm,egm-resolve): Promoted the non-empty visible chat requirement to globalInteraction Invariants, required structured question gates to run as the sole tool call in their model response step (after earlier preparatory file writes/reads in the same user turn complete, preventing concurrent[write_file, ask_question]tool-batching races), and explicitly exemptedegm-resolveStep 4 andInteractive breakdownso their 5-option / 3-option plain-chat menus remain zero-tool turns.Contribution Type
Track: Feature / Architectural Change -- Stage 2: Implementation PR
Tracked via maintainer refactor issue #371 (
pr-ready); no design document added underdocs/designs/.Verification
origin/main(0997679, preserving#409.NET / C#bootstrap additions and#427ruff/actionlint checks). Offline Python unit checks (python3 -m pytest tests/unit -k "not test_ci_workflow and not test_scope_advisory" -q): 378 passed, 3 skipped (Node-dependent tests,lint, andsite-validationverified in GitHub Actions).SKILL.md->resources/execution_instructions.md), all 43skills/*/test_scenarios.yamlscenarios A/B verified againstorigin/main(0regressions), 9 live tool-using & nested-subagent E2E workflows across all 7 skills, plus 4/4 post-patch live subagent verification probes (egm-bootstraphidden-dir ignore,egm-reviewOption 1max 1 re-audit+ direct-dispatch standard mapping/retry, andegm-draftDesign-Only Scratchpad & template checks).21/21interactiveask_questionmodals + plain-chat guards acrossegm-draftandegm-review), interactive CLI TUI modal navigation, and10/10side-by-side multi-turn gate evaluation scenarios againstorigin/main.Human Accountability Affirmation