Agent Skills

Canon — Orchestrator

Runs Chronos agent presets, aggregates reports, validates cross-references between L1-L3.

Presets

Preset	Agents	Description
minimal	Censor	Duplicates + broken links only
standard	Censor + Dewey + Canon	Check + classify + cross-ref
full	All 5 agents	Complete Pantheon

What it does

1. Receives preset (minimal/standard/full)
2. Determines required agents
3. Executes agents in sequence
4. Aggregates findings into unified report
5. Validates L1→L2→L3 cross-references (contracts match docs)

Usage

# Via Chronos orchestrator
chronos --path . --preset full

# Direct invocation
python -m chronos.agents.canon --path . --preset full

Cross-Reference Validation

Canon checks consistency between levels:

• L1→L2: OpenAPI endpoints documented in PRD/FEATURES
• L2→L3: Architecture decisions trace to PRD requirements
• L3→L1: ADRs reference contract versions

Output Format

[info] preset: full
[info] agents_run: censor, dewey, veles, canon
[warning] cross_ref: GET /users in openapi.yaml (v2.1) but PRD.md shows v2.0
[critical] cross_ref: ADR-005 references deprecated endpoint removed in v2.1

Triggers

• run full audit / full audit
• chronos full / chronos --preset full
• cross-reference validation / cross ref check
• canon

Integration

Loaded by Chronos as part of standard and full presets. Canon is both an agent and the preset orchestrator.

Boundaries

• Do not use as a standalone documentation analyzer; use the focused chronos, censor, dewey, or veles skill when only one audit is needed.

Inputs and decision rules

Input is a repository path and one preset. Start with minimal when the user asks for a narrow check; use standard for a PR and full for a release audit. Never infer that a missing document is a defect without showing the detected project level and the configured requirement.

Output gate

A run is complete only when the selected agents, issue count, preset, project path, and output format are visible in the report. For CI, use JSON and --fail-on; inspect at least one finding’s file, category, severity, and fix. A zero-issue report is not proof that unscanned file types or unavailable external systems were checked.

This site is open source. View this page on GitHub.