shards-foundation/shards-mcp

Safely enrol, validate and offboard a least-privileged Shards-MCP agent through host-local admin controls.

Triage a failing Shards-MCP GitHub Actions job or local test, reproduce the smallest fault, repair it with a regression and rerun the relevant matrix. Use for failed CI checks, dependency problems, typecheck failures and review-blocking build errors.

Configure and verify MCP connectivity for Claude Desktop, Claude Code, ChatGPT, Codex app or CLI, Cursor app or CLI, and other protocol clients. Use when implementing real client onboarding, choosing transport, testing OAuth scopes or troubleshooting client setup.

Read-only Shards-MCP transport and permission diagnosis for a separately enrolled agent.

First-run Shards-MCP onboarding for a new human contributor or AI coding agent. Use to inspect the repository, discover skills, verify local prerequisites and complete a safe, read-only first workflow without credentials or hosted-service assumptions.

Contain credential, task-lease and data-integrity incidents without replaying uncertain external actions.

Prepare evidence and platform-specific submissions for MCP directories and app marketplaces without implying certification or listing approval. Use for ChatGPT, Claude, GitHub, Microsoft, Google Workspace, Vercel, Oracle or third-party MCP/Discord directories.

Verify source/PyPI release or hosted-service gates with evidence and human approvals; prevent false production claims.

Bounded adversarial review of Shards-MCP trust boundaries and CI evidence; never assert certification without deployment tests.

Design, write, test and safely register a new repeatable Shards skill following the Agent Skills specification. Use when adding workflows, scripts, playbooks, reference documents, or extending the skills catalog and agent onboarding instructions.

Submit, claim, report and independently verify a bounded Shards-MCP task without blind external-effect retries.