mcp-server

Validate review findings about an implementation plan against the codebase and repo rules, fix the valid ones in the plan, and write a feedback file noting what changed and what was rejected (and why). For plan-review findings, not code-review findings.

Expert review of an implementation plan against a GitHub issue/enhancement description (provided as a local file or a GitHub issue URL) and the current repository codebase. Use when asked to critique a plan for correctness, completeness, codebase alignment, risks, and test/rollout readiness (do not implement).

Manually review a feedback file produced after implementation-plan review findings were addressed. Use when the user asks to read a findings-feedback file and verify that the plan changes correctly close, correctly reject, or fail to close the original implementation-plan-review findings, while checking for newly introduced or newly exposed plan problems. Writes only new findings to the plan-id findings directory and replies with the output file path or a concise no-new-findings status.

Use this skill whenever the agent needs to deploy a smart contract to Sepolia testnet via Foundry (`forge create` or `forge script --broadcast`), routed through the Blockscout PRO API's JSON-RPC gateway instead of a third-party RPC provider. Trigger on requests like "deploy this contract to Sepolia", "deploy via Foundry", "broadcast this Foundry deployment script", or any `forge create`/`forge script` invocation that needs a signer and an RPC endpoint for Sepolia (chain id 11155111) in this project. Also consult it to find which local Foundry keystore to sign with — this project stores test signing keys as encrypted keystores under `.ai/wallet/`, never as a plain private key in `.env`. Do not use this for reads (balances, transaction history, contract info) — use the connected Blockscout MCP tools for those directly; this skill is specifically about broadcasting a deployment.

Enter the worktree of a GitHub PR, validate the code-review comments raised on it against what the code actually does and the repo rules, fix the valid ones directly in the PR's source/tests, and report in chat — a brief note per closed comment and a detailed plain-language explanation of every comment left unclosed and why. For code-review comments on a PR. Invoke manually.

TRIGGER CONDITIONS - invoke this skill when ANY of these apply: User-triggered (explicit request): - User mentions "specification", "SPEC", "SPEC.md", "spec requirements" - User asks to "verify with spec", "confirm with specification", "check requirements" - User requests "authoritative requirements" or "source of truth" Agent self-triggered (during reasoning): - Before implementing any feature or making code changes - When verifying technical requirements or constraints - When making architectural or design decisions - When you need unbiased guidance not influenced by current implementation - When encountering conflicting information and need ground truth PURPOSE: Consults SPEC.md through an isolated subagent that ONLY reads the specification, never implementation code. This ensures guidance is based on authoritative requirements, not potentially incorrect existing code.

Describe very high level changes required to implement discussed functionality (functionality, test, docs)

Generate a structured bug report for GitHub issues based on conversation context about bugs, errors, or broken functionality

Generate a structured improvement/enhancement proposal for GitHub issues based on conversation context

Publish a GitHub issue from a markdown file in .ai/gh_issues/ created by gh-issue-bug or gh-issue-improvement skills

Use this skill whenever you are about to run any write `gh` CLI command — `gh pr edit`, `gh pr create`, `gh issue edit`, `gh issue create`, `gh pr comment`, `gh issue comment`, `gh pr close`, `gh pr merge`, `gh release create`, or similar. High-level `gh` commands sometimes fail with "missing required scopes" auth errors even though the underlying GitHub REST API works fine with the same token. This skill makes you automatically fall back to `gh api` instead of stopping to ask the user to re-authenticate. Always invoke before running any mutating `gh` command.

Export a detailed phased implementation plan for a GitHub issue to a file, reading all applicable rules from .cursor/rules before planning

Critically review a GitHub issue plus a "described changes" overview written by a developer — open the referenced code, verify the overview's claims against what the code actually does, and report gaps, inconsistencies, and likely mistakes — without modifying any files. Invoke manually.

Use this skill WHENEVER you are about to run the integration tests of this project — the whole suite, a single module, or one test. This includes any time you would otherwise type `pytest -m integration`, `uv run pytest -m integration`, run anything under `tests/integration/`, or want to check live Blockscout/Chainscout/BENS connectivity. Integration tests make real network calls with no hard HTTP timeout, so one unresponsive endpoint can hang a plain pytest run for an unbounded time and block you. This skill runs each test in an isolated subprocess with a per-test timeout, kills any hung test, and always returns a bounded report flagging which tests timed out or are slow. Prefer it over invoking pytest directly for integration tests.

Use when user asks to "run tool as function", "test directly", or "call standalone" (outside MCP server). Essential for correct mock Context setup (MagicMock + AsyncMock). Finds integration tests with real test data.

Use to implement a structured implementation plan (e.g. temp/impl_plans/issue-*.md) phase by phase. Trigger when the user asks to implement, execute, realize, or "do" an implementation-plan file. Orchestrates a phase-developer subagent and a plan-correspondence-verifier subagent for each phase, commits each verified phase, then walks the Final Checklist. Invoke explicitly — it makes commits and spawns many subagents.

Reads the content of a file located in the temporary directory (temp/) which is otherwise ignored by git.