| name | meritranker-project-skills |
| description | Orchestrates MeritRanker Tutor repo-local skills and AI development roles from skills/. Reads AGENTS.md, core rules, feature context, and applies roles contextually for quality, bug-free code. Use when working in this repository, implementing features, fixing bugs, reviewing changes, or when the user wants project-standard quality. |
MeritRanker Project Skills & Roles
Apply all relevant repo guidance under skills/ on every task. Run roles contextually — not every role on every change, but never skip required evidence or gates.
Always read first
- AGENTS.md — repo boundaries, hard rules, role sequence
- skills-index.md — linked index of every
skills/ file
- Matching skills/features/.md for the feature in scope
Default knowledge load (by task type)
| Task type | Core files (read before acting) | Feature file |
|---|
| Any code change | coding-standards.md, architecture-principles.md, testing-and-debugging.md | Yes |
| Graph / workflow | + langgraph-patterns.md, pydantic-schemas.md | Yes |
| Schema / API contract | + pydantic-schemas.md, integration-boundaries.md | Yes |
| Deploy / AgentCore | + agentcore-runtime.md | Yes |
| LLM / prompts / tools | + langgraph-patterns.md, security-and-privacy.md, performance-and-scalability.md | Yes |
| External integration | + integration-boundaries.md, security-and-privacy.md | Yes |
| Docs-only | documentation-rules.md | Yes |
Paths are under skills/core/ unless noted.
Contextual role workflow
Pick the minimum role set that satisfies the task. Escalate when scope, risk, or ambiguity grows.
Role selection matrix
| User intent | Active roles (in order) | Planning required? |
|---|
| New feature / major capability | PM → BA → Solution Architect → AI Solution Architect → Engineer → (Architect re-check) → QA + Security + Performance + Docs → Release Gatekeeper | Yes — no code until 1–4 align |
| Prompt / generator / classifier wording changes | Prompt Engineer → Python Agent Engineer (if code) → QA | Plan or task brief for prompt scope |
| Approved implementation only | Python Agent Engineer → QA (+ Security/Performance if applicable) → Docs Maintainer | Plan must exist |
| Bug fix | Engineer (use bugfix template) → QA → Security if logs/auth/secrets touched | Light — document in bugfix report |
| Small fix (typo, comment, single-line) | Engineer hat + make check | No |
| Code / PR review | QA Reviewer; add Security + Performance for risky areas | No |
| Architecture question | Solution Architect (+ AI Solution Architect if graphs/LLM) | No |
| Release / merge readiness | All prior evidence → Release Gatekeeper | Evidence from reviews |
| Documentation sync | Documentation Maintainer | No |
Role files (read the active role guide before acting)
Role boundaries: skills/roles/README.md
Output templates (use when producing role deliverables)
Match template to role — see skills/templates/README.md. Do not delete template sections; use N/A, [NOT VERIFIED], or [BLOCKER] as needed.
Completion gates (every code change)
- [ ] Read AGENTS.md + relevant skills/core/*.md + skills/features/<feature>.md
- [ ] Acted in correct role(s); escalated if scope changed
- [ ] Code only in app/; no secrets; schemas unchanged unless approved
- [ ] Tests added/updated for behavior changes
- [ ] make check passes
- [ ] agentcore validate passes (if agentcore/ or deploy touched)
- [ ] skills/features/<feature>.md updated (if feature code changed)
- [ ] Unverified items labeled [NOT VERIFIED] — never claim certainty without evidence
Hard rules (never violate)
From AGENTS.md — summarize only; full text is authoritative:
- No FastAPI; AgentCore provides HTTP
- No infra (DynamoDB, Redis, S3, KB) unless explicitly requested
- No real LLM calls in demo/foundation unless explicitly requested
agentcore/ = config only; all Python in app/- Pydantic v2 at API boundaries; LangGraph for workflows
- Do not break public schemas without explicit approval + tests
When to expand scope
Stop and run additional planning roles (PM → BA → Architects) if:
- Requirements are unclear or acceptance criteria missing
- Public schema or graph topology would change
- New external integration or infrastructure is needed
- Security, AI, or performance risk is unclear
Additional resources
