feat: debug-agent Claude Code plugin (debugger + Python/Go/Node skills + agents)#7
Merged
Conversation
…evidence-first, dependency-hygiene)
…hon/TS type examples
…tive rate is harness ceiling)
…dd marketplace description
…ompts Dogfooding the plugin's own agents on this repo surfaced a systemic gap: all four experts default to confident source-reading on review/audit tasks (no live failure to reproduce) without labeling confidence. node-expert shipped a HIGH-severity false positive that a single parser run disproved, and its proposed fix was behaviorally identical to the existing code. Add a 'live-failure vs static review' mode to the shared evidence-first discipline and an in-body review/audit clause to python/go/node-expert and architect: label findings RUNTIME-VERIFIED vs INSPECTION-ONLY, prove or offer a repro for anything reproducible, and separate 'breaks today' from 'latent under a future/edge runtime'. Kept in agent bodies because dispatched agents were observed skipping reference loads.
Child-delegating adapters (vscode-js-debug) run the program in a child session, so exception filters set on the parent at launch never bind. Breakpoints were already stashed and replayed on the child; do the same for exception filters. Without this, `--break-on-exception` was silently dropped for Node. Stash `_exception_filters` at launch and replay them in `_on_start_debugging` alongside breakpoints.
…E usage guidance node-expert: make the top-level evidence stance unconditional (mirrors python-expert rule #3) so the RUNTIME-VERIFIED/INSPECTION-ONLY labeling discipline applies to review/audit tasks, not just crash-fix flows. README: add usage guidance (architect delegation cliff, review-vs-debug expectations, opus override for hard single-language tasks).
# Conflicts: # CHANGELOG.md
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.
Packages the
dbgaevidence-first debugger plus consolidated language skills and specialist agents as a distributable Claude Code plugin, giving a full design → develop → debug → verify → clean-up workflow for Python, Go, and Node/TypeScript.Implements the spec/plan in
docs/superpowers/(borrow-and-refine from MIT sources wshobson/agents + VoltAgent).What's in it
.claude-plugin/marketplace.json(namedbga, source./plugin) +plugin/.claude-plugin/plugin.json(namedebug-agent, v0.1.0, MIT). README documents both install paths + thedbga/debug-agent/debug_agentname glossary.plugin/skills/):debug-agent(moved from repo root — 5 refs updated), plus new slim-indexpython,go,nodeskills (all SKILL.md < 500 words) routing to language-specific references. Language-invariant discipline authored once in_shared/(clean-code, evidence-first, dependency-hygiene) and cross-referenced by name.plugin/agents/):architect(opus orchestrator, authored fresh) +python-expert/go-expert/node-expert(sonnet), each embedding the Evidence-First block./debug-agent:setup(uv → pipx → pip installer, toolchain readiness report).references/agent-teams.md(experimental parallel-debugging mode),THIRD_PARTY_NOTICES.md(verbatim MIT texts + per-file SHAs), CLAUDE.md "Python-only" line corrected to the multi-language reality.Verification
claude plugin validate ./pluginpasses; all 4 skills + 4 agents discovered and valid.npx skills add <repo> --skill python|go|node|debug-agenteach install standalone (resolved via marketplace.json — no--full-depthneeded; Phase 0 verified empirically against skills CLI v1.5.0).quick_validatepasses for all 4 skills.dbga sessionshapes + dep-hygiene framing + Python/TS type examples.select()-on-pipe bug → WinError 10038); needs WSL — documented indocs/superpowers/evals/RESULTS.md.Follow-ups (not in this PR)
claude --plugin-dir ./plugin/help+/agentslisting and the architect-as-main-thread e2e are interactive checks to run post-merge.