great_cto

v2.25.0 — 2026-05-28

agent-workflows patterns: triage gate, structured findings, hand-off rules

Inspired by franklioxygen/agent-workflows.

Triage Gate — depth-based pipeline selection

• Added Triage Gate section to CLAUDE.md request classifier — classifies SIMPLE CODE (Tiny/Small/Medium) and COMPLEX CODE (Small/Medium/Large) before pipeline runs
• Added Escalation guard: if scope grows during work → stop and reclassify; do not continue with under-powered process

Reproduction Requirement — bug-fix gate

• Added Reproduction Requirement to agents/coordinator.md — before any implementation agent edits code for a bug fix, must establish: failing automated test, failing repro command, or explicit infeasibility note. "It would take effort" is NOT infeasible.
• Blocks implementation packet until repro packet completes

Baseline Establishment — refactoring gate

• Added Baseline Establishment to agents/coordinator.md — run full validation set (tests + lint + build) before refactoring edits; record pass/fail. Separates new regressions from pre-existing failures.

Structured Findings Format — review output standard

• Added Structured Findings Format to skills/great_cto/SKILL.md — Critical/Major/Minor/Nit severity tiers with Location/Problem/Why it matters/Recommended fix/Status fields
• Added mandatory Summary block (APPROVED/BLOCKED verdict) to all review output
• Critical + Major findings block merge and gate:ship

Workflow Hand-off Rules

• Added explicit hand-off transition table to agents/coordinator.md — INCIDENT→BUG-FIX, BUG-FIX→Feature, SIMPLE→COMPLEX, Cleanup→Refactoring, etc.
• Silent workflow class switches are explicitly prohibited; always notify CTO with reason

Per-workflow Safety Rules

• Added shared/safety-rules.md — five safety rule variants: Standard Coding, Review-Only, Behavior-Preserving, Cleanup, Incident Response
• Each rule matched to the workflow class that should use it

Scope Escalation Guards

• Added scope escalation guard to coordinator anti-patterns: "scope creep is low risk" → stop + reclassify

Minimal Loading Discipline

• Added per-class loading rules to skills/great_cto/memory-index.md — load only what the active workflow needs; per-class breakdown (QUESTION, SIMPLE CODE Tiny, INCIDENT)

ag-kit patterns: coordinator, skillify, memory index (v2.24.1 content shipped in v2.25.0)

• Added agents/coordinator.md — multi-agent orchestrator with DECOMPOSE→CLASSIFY→DISPATCH→MONITOR→SYNTHESIZE→VERIFY lifecycle, Work Packet List format, Fork vs Spawn semantics

• Added commands/skillify.md — interactive 6-question interview to capture repeating patterns as SKILL.md files

• Added skills/great_cto/memory-index.md — 200-line cross-session knowledge index template with 3-level compression protocol

• Added Request classifier section to CLAUDE.md — 8-class routing (QUESTION/SURVEY/SIMPLE CODE/COMPLEX CODE/DESIGN/SLASH CMD/INCIDENT/COORDINATE)

• Updated dispatch semantics in skills/great_cto/SKILL.md — Fork vs Spawn, Never Delegate Understanding, concurrency safety

• Updated context compression protocol in commands/save.md — micro-compact, phase summary, checkpoint levels

• Updated validation hierarchy reference in commands/doctor.md — 7-layer priority (Security → Schema → Tests → Lint → Performance → UX/A11y → SEO)

• Updated all 13 SKILL.md files — added effort: low | medium | high frontmatter field

• Add one bullet per shipped feature.

• Cite ADRs introduced (if any).

• Mention test counts and opt-out flags.