Menu
Use LangSmith CLI to inspect LLM traces/runs for Career Forge debugging. Use when debugging diagnosis interview, forge SSE, mentor flows, or any upstream LLM execution.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Use when debugging local failures, failed to fetch, API errors, Docker stack issues. Always apply for bug reports and runtime errors in Career Forge.
Sync claude-design-docs with implemented UI after paradigm changes. Use when finishing UI work, when asked to "sync UI product docs", or when prototype vs code diverged.
Mandatory pre-merge triple gate for Career Forge — code evaluator, Playwright MCP, agent verify. Use before every merge.
Gate C — run make agent-verify for API + Postgres proof before merge. Extend scripts/agent-verify.sh for new behavior.
Basé sur la classification professionnelle SOC
| name | langsmith-inspect |
| description | Use LangSmith CLI to inspect LLM traces/runs for Career Forge debugging. Use when debugging diagnosis interview, forge SSE, mentor flows, or any upstream LLM execution. |
Inspect LangSmith traces and runs for Career Forge LLM flows (GraphExecutor, LangGraph).
Helper script: scripts/langsmith-env.sh
Always load from repo .env before any CLI call. Never print or commit API keys.
cd /path/to/HB01-2026_soft-push
set -a && source .env && set +a
Or use the helper (recommended):
./scripts/langsmith-env.sh trace list --project "$LANGSMITH_PROJECT" --last-n-minutes 60
Default project: $LANGSMITH_PROJECT (career-forge from .env.example).
If auth fails, verify LANGSMITH_API_KEY is set in .env (not .env.example). Do not ask the user for the key when .env exists — read it via the helper.
CLI install (if missing): curl -fsSL https://cli.langsmith.com/install.sh | sh → ~/.local/bin/langsmith
Recent traces (human-readable):
./scripts/langsmith-env.sh trace list --project "$LANGSMITH_PROJECT" --last-n-minutes 60 --format pretty
Full trace hierarchy (GraphExecutor / nested runs):
./scripts/langsmith-env.sh trace get <trace-id> --project "$LANGSMITH_PROJECT" --full --show-hierarchy --format pretty
Recent LLM runs with metadata:
./scripts/langsmith-env.sh run list --project "$LANGSMITH_PROJECT" --run-type llm --include-metadata --last-n-minutes 60
Single run detail:
./scripts/langsmith-env.sh run get <run-id> --full
Filters:
| Filter | Flag |
|---|---|
| Errors only | --error |
| Slow traces (≥5s) | --min-latency 5 |
| By graph/name | --name diagnosis_interview |
--format pretty — tables, trees, highlighted JSON (default; use for human review)--format json — machine-readable (pipe to jq, scripts, agents)Search traces/runs by --name or tags when debugging:
| Graph / flow | Typical --name |
|---|---|
| Adaptive diagnosis | diagnosis_interview |
| Live Roadmap Forge | forge, roadmap_forge |
| Mentor chat | mentor |
| Mastery validation | validation |
Prefer trace get … --show-hierarchy for GraphExecutor flows — nested LLM/tool nodes appear as a tree.
make up, hit the API route or UI flow)../scripts/langsmith-env.sh — confirm project + API key status.--last-n-minutes 60); add --error or --min-latency 5 if needed.GraphRun / run_id with LangSmith trace ID when available../scripts/langsmith-env.sh
Prints repo path, project, key status (length only), and CLI version.