Janus Analyzer Skill

Implementation Rules

• Keep analyzer names in kebab-case and register outputs in Core/analyzer_registry.py.
• For a new analyzer, update all runtime registration points:
  • Core/analyzer_registry.py
  • Config/analyzers.yml
  • janus.py ANALYZER_FUNCTIONS
  • janus.py run_analyze()'s explicit analyzer dispatch branch
• Preserve the canonical task/result event model Janus already uses.
• Treat weaker source fidelity as a capability boundary, not as a reason to flatten stronger sources.
• Treat Ghostwriter result state as unreliable unless the source schema actually adds trustworthy status fields.
• If a metric depends on success/error truth, say explicitly which Janus-supported sources can support it today.
• If only heuristics need adjustment for known commands, prefer Config/analyzer_registry.yml over hard-coding special cases.
• Always update Core/html_output.py when an analyzer's output schema changes. Find the _render_<analyzer_name> function and update any .get("key") references that no longer match. Silently empty tables are the failure mode.
• After adding a new analyzer, run it once through the real CLI/container path, not just direct Python imports, to catch missing run_analyze() registration.
• Parser robustness guardrails now exist and should be preserved:
  • normalize_timestamp() accepts ISO and epoch values; malformed timestamps should be handled intentionally (skip with counters or raise early).
  • load_events() can validate schema and warns on unknown event types.
  • run_merge() remaps missing or duplicate operation_id values to deterministic unique IDs to avoid cross-operation key collisions.

What Good Output Looks Like

When answering the user, produce:

• The metric or question being answered
• The source scope and any fidelity limits
• Whether Janus already supports it or needs code changes
• If code changes are needed, the minimal files to touch
• Which bundled source reference informed the source-specific guidance

References

• docs/architecture.md
• docs/partial-data.md
• docs/ghostwriter-api-reference.md