docs(artifacts): compress skills/agents ~40%; fix 8 defects from review#390
Merged
Conversation
Shipped consumer artifacts (skills/ + agents/) load into an LLM's context at runtime, so verbosity is a direct token/perf tax. Compress all 5 files ~40% (1128 -> 685 lines) by collapsing the Decision-Framework / Decision-tree / Workflow-Patterns triple-listing into one canonical table, cutting generic Glob/Grep boilerplate, and deduping repeated explanations. Each file stays a standalone manual (skills and agents load via different mechanisms; one surface deployed per project). A parallel 3-reviewer pass against source (jrag.py, mcp_v2.py, java_ontology.py, ladybug_queries.py) found 8 factual defects; only 1 was introduced by the compression, the rest were pre-existing and carried forward: - CLI: `routes`/`clients` -> real `http-routes`/`http-clients` subcommands - CLI: `not_found` exits 0, not 2 (only `error` exits 2) - CLI: `impact` is INJECTS/IMPLEMENTS/EXTENDS only (no CALLS) - MCP+CLI: route `framework` also accepts kafka/rabbitmq/jms/stream/feign - MCP: restored "often" hedge on empty-string filter behavior (compression-introduced) - CLI: added missing `decompose` command; `hierarchy` walks both directions - CLI: `--service`/`--module` are resolve-time filters on inspect/callers install_data copies re-synced. The MCP skill keeps its 6 required headings, 5 tool refs, and valid kinds/edges (test-pinned). Verified: test_agent_skills_static (21), test_install_data_sync (4), test_installer_surface (13) = 38 passed. Co-Authored-By: Claude <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
skills/andagents/are shipped verbatim to consumers and load into an LLM's context at runtime — verbosity is a direct token/perf tax. This PR compresses all 5 files ~40% (1128 → 685 lines) and fixes 8 factual defects surfaced by a source-verified review.Compression (Firm ~40%)
Applied uniformly:
skills/README.md(not shipped)skills/explore-codebase/SKILL.md(MCP)skills/explore-codebase-cli/SKILL.md(CLI)agents/explorer-rag-enhanced.md(MCP)agents/explorer-rag-cli.md(CLI)8 factual defects fixed
Found by a 3-reviewer pass that verified every claim against
jrag.py,mcp_v2.py,java_ontology.py,ladybug_queries.py. Only 1 was introduced by the compression; the other 7 were pre-existing and carried forward.routes/clients→ realhttp-routes/http-clientssubcommandsnot_foundexits 0, not 2 (onlyerrorexits 2)impactisINJECTS/IMPLEMENTS/EXTENDSonly (noCALLS)frameworkalso acceptskafka/rabbitmq/jms/stream/feigndecomposecommandhierarchywalks both directions (parents + children)--service/--moduleare resolve-time filters oninspect/callersTest-structure preserved
The MCP skill is statically pinned by
test_agent_skills_static.py. Verified intact: frontmatter, all 6 required headings (Workflow Patterns,Decision Framework,Recovery Playbook,Edge taxonomy,Tool Inventory,Forced reasoning preamble), all 5 tool refs, and valid kinds/edges/directions.Verification
Follow-ups (not in this PR)
docs/AGENT-GUIDE.md:248has the same route-frameworks gap as old Add per-PR Cursor task prompts for Tier 1 completion #4 (pre-existing). Fixing Add per-PR Cursor task prompts for Tier 1 completion #4 in the skill diverges from that "source of truth" doc — can sync it in a follow-up if wanted.--limit "default 20; 10 for fan-out"mirrorsjrag.py's own help string, but no command actually defaults to 10 — a bug in the CLI help text itself, left untouched here.🤖 Generated with Claude Code