code-copilot-team

Multi-agent delegation rules, three-phase workflow (Plan/Build/Review), model selection, collaboration gates, and Ralph Loop integration.

Autonomous build driver after SDD spec approval: phase-scoped headless build sessions, driver-owned commits, cross-provider review gates, milestone pauses, and fail-closed escalation — under explicit opt-in autonomy profiles.

When and how to ask clarifying questions before implementing. Data model review gate and ambiguity resolution rules.

Quality gates, prohibited patterns, verification discipline, and self-audit rules for all code generation and review sessions.

Cross-copilot portable conventions: read before write, minimal changes, git discipline, project structure, and priority rules.

Copyright header rules for generated source files. Applies the company name from providers.toml to new files.

Derive a unique, domain-fit design direction and enforce design-token + anti-slop discipline. Read/author DESIGN.md before building any UI; override framework defaults so output is bespoke by construction, not generic.

Environment variable patterns, config file validation, and setup verification for new project scaffolding.

Infrastructure artifact verification: Docker builds, CI workflows, and launcher flags must be executed, not just syntax-checked.

Test integration points early. Verify cross-service contracts, API boundaries, and data flow before declaring done.

MemKernel persistent memory protocol. Self-guarding: activates only when the memkernel MCP server is configured.

Optional guidance for sessions using Claude Opus 4.7: adaptive thinking, xhigh effort, prompt caching, /btw side questions.

Origin-confirmation circuit breaker: machine-checkable origin frontmatter, alignment-check protocol, three gates (plan-approval, build-entry, phase-complete), and interactive escalation on deviation.

Phase transition rules, post-phase verification steps, peer review validation, and commit-gate approval flow.

Cross-provider peer review protocol: session flags, review flow, collaboration artifacts, fail-closed enforcement, and CI validation.

Single-agent autonomous iteration loop: PRD-driven, self-verifying, with configurable iteration limits and completion criteria.

Agent-driven peer review loop: structured findings, dispositions, circuit breakers, commit strategies, sandbox isolation, and plan-phase advisory mode.

Non-negotiable safety constraints: destructive action guards, blocked operations, secrets policy, password storage, input validation, and dependency review.

SDD specification protocol: risk-based spec_mode classification, required sections per mode, plan approval gate, and artifact directory conventions.

Stack version pinning and dependency compatibility guards. Prevents silent version drift across project boundaries.

Build team lead efficiency rules: limit sub-agents, polling discipline, no redundant re-work, parallel task launching.

Token economy rules: diff-over-rewrite, context compression, avoid verbose re-explanations, minimize round-trips.

Closed visual-review loop for generated UI: render the running app, screenshot per breakpoint, run an axe-core a11y gate, critique against the DESIGN.md rubric, triage findings, and iterate to a quality bar with a hard cap. Drives the shippable harness/ runner.

Wiki-first query convention: consult knowledge/wiki/index.md and the linked pages BEFORE re-reading raw sources for a project topic. The wiki is the canonical project memory layer.

Executes an approved plan using the Ralph Loop pattern — iterate through tasks sequentially, verify after each step. The only phase that writes code.

Runs end-to-end verification of the project. Executes test suite, type checker, linter, and dev server. Reports pass/fail with specific failure details.

Produces implementation plans with files, interfaces, test strategy, and task breakdown. No code changes.

Explores codebase, reads docs, searches the web. No code changes. Returns a research summary with paths, patterns, and risks.

Holistic review of all changes — correctness, consistency, style, test coverage. Runs tests, checks console, verifies integration. Returns pass/fail report.