From context-probe-plugin
Run `context-probe` measurements and analyze the results after repo inputs exist. Executes `score.compute`, `report.generate`, `gate.evaluate`, `review.list_unknowns`, and deeper `doc.extract_*` `trace.*` `history.*` or `review.resolve` only when needed. Use when asked to measure a repo, analyze a result, inspect unknowns, or continue after YAML inputs are ready.
How this skill is triggered — by the user, by Claude, or both
Slash command
/context-probe-plugin:context-probe-analysis-workflowThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Run `context-probe` after curated inputs already exist.
Run context-probe after curated inputs already exist.
Respond in English while this skill is active unless the user explicitly asks for another language.
If the repo does not have maintained YAML inputs yet, switch to context-probe-inputs-workflow first.
Read references/command-recipes.md when you need exact command shapes. Read references/interpret-results.md immediately after any run. Read references/investigate-correctness.md only when the result looks suspicious or too weak to trust. Read references/final-summary.md before closing.
Place the user in one of these paths before running anything:
domain_design measurementarchitecture_design measurementRun the narrowest command that answers the request:
score.computereport.generategate.evaluatereview.list_unknownsreview.resolve: use only when review items and explicit human resolutions already existRead outputs in this order:
statusresultconfidenceunknownsdiagnosticsprovenanceSeparate the result into:
If results look suspicious, incomplete, or lower-confidence than expected, inspect likely causes before changing curated YAML:
review.list_unknowns to see unresolved gapsdoc.extract_* to inspect extracted terms, rules, or invariantstrace.link_terms or trace.link_model_to_code to inspect linking evidencehistory.* to inspect locality and co-change signalsreview.resolve only to apply explicit human review decisions, then rerun the affected extraction or score pathEnd with a concise summary that states:
Do not turn every run into an issue proposal. Finish with the measured state first.
doc.extract_* for a first measurement unless direct evidence inspection is the task.--model and --constraints.status: ok as proof that the result is complete or high confidence.unknowns, diagnostics, or provenance.npx claudepluginhub caphtech/context-probe --plugin context-probe-pluginRuns an interactive analysis lane for traceable investigation questions with enter, continue, end, and report commands.
Verifies data science analysis results for reproducibility and completion, using guardrails to gate tool usage until approval.
Investigates codebases in parallel, producing a layered Investigation Brief with core findings and optional depth modules. Use before any codebase task.