Skip to content

Latest commit

 

History

History
84 lines (60 loc) · 2.97 KB

agents.md

File metadata and controls

84 lines (60 loc) · 2.97 KB

CCO Agents

Specialized subagents for analysis, fixes, and research.

Overview

Agent Purpose Model Isolation Tools
cco-agent-analyze Codebase analysis with severity scoring Haiku/Sonnet/Opus* — Glob, Read, Grep, Bash
cco-agent-apply Verified write operations + accounting Inherited — Grep, Read, Glob, Bash, Edit, Write, NotebookEdit, AskUserQuestion
cco-agent-research Multi-source research with CRAAP+ scoring Haiku — WebSearch, WebFetch, Read, Grep, Glob

Model rationale: Per CCO Rules: Model Routing. Agent frontmatter defaults to Haiku. Skills override via Task tool's model parameter: auto → Haiku, review → Sonnet, CRITICAL escalation → Opus. Apply agent inherits the session model.

When to Use

Need information?
├── Single URL/fact → WebSearch/WebFetch
└── Multiple sources/verification → cco-agent-research

Need analysis?
├── Find file/pattern → Glob/Grep/Read
└── Structured findings/metrics → cco-agent-analyze

Need changes?
├── Single file edit → Edit/Write
└── Multiple files/verification → cco-agent-apply

cco-agent-analyze

Read-only analysis agent. Returns structured JSON with findings, scores, and metrics. No worktree isolation — read-only parallel access is safe.

  • 9 optimize scopes (97 checks) — see agents/cco-agent-analyze.md
  • 8 review scopes (92 checks) — see agents/cco-agent-analyze.md
  • 4 audit scopes (40 checks) — see agents/cco-agent-analyze.md
  • Context-aware filtering by project type, cross-scope dedup, negative evidence checks
  • Standardized score calculation formula (penalty-based with severity caps)
  • Per CCO Rules: Confidence Scoring, Severity Levels, Skip Patterns, Fix Quality

cco-agent-apply

Write agent. Applies fixes with verification and cascade handling.

  • Batch edits with parallel execution
  • Post-change verification (lint/type/test)
  • Cascade fixing (max 3 iterations)
  • Per CCO Rules: Accounting, Auto Mode
  • Educational output (why/avoid/prefer)
  • Docs scope for documentation generation

cco-agent-research

Research agent. Multi-source with CRAAP+ reliability scoring.

  • 6 scopes: local, search, analyze, synthesize, full, dependency
  • T1-T6 source tiers with modifiers
  • Contradiction detection and resolution
  • Saturation detection (auto-stop)
  • Dependency mode with registry API integration

Agent Selection by Skill

Skill Analyze Apply Research
/cco-optimize security, hygiene, types, etc. Yes dependency
/cco-align architecture, patterns, etc. Yes dependency
/cco-commit (quality gates only) No No
/cco-research - No full
/cco-docs docs scope Yes No
/cco-blueprint all scopes (optimize + review + audit) Yes No
/cco-pr - No No
/cco-update - No No

Back to README