Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/kunwarvivek/mcp-github-project-manager/agents-mdgit clone --depth 1 https://github.com/kunwarVivek/mcp-github-project-managerWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.03572 | $0.03572 |
| Opus 5 | $0.01786 | $0.01786 |
| Sonnet 5 | $0.00714 | $0.00714 |
| Haiku 4.5 | $0.00357 | $0.00357 |
Grade A, and why
mcp-github-project-manager AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
Copies of this mod
1 near-identical copy found in the catalogue:
- mcp-github-project-manager CLAUDE.md — 94% identical, 28 lines differ
How it starts
The opening of the file, as written. The whole thing — 238 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Commands
| Task | Command |
|---|---|
| Build | npm run build (tsc → fix-imports.js → chmod) |
| Dev (watch) | npm run dev (tsx — ts-node cannot load under TS 7) |
| Unit tests | npm test · core only npm run test:core · AI npm run test:ai |
| E2E (mock) | npm run test:e2e · tools npm run test:e2e:tools |
| E2E (live API) | E2E_REAL_API=true npm run test:e2e:tools:real |
| Coverage | npm run test:coverage |
| Lint / format | npm run lint (Biome) · npm run format |
| Inspect MCP server | npm run inspect |
Architecture (DDD, MCP stdio server)
src/index.ts— entry; boots anMcpServer(@modelcontextprotocol/serverv2, 2026-07-28 spec) viaserveStdio, which negotiates the protocol revision per connection. Each tool is registered withregisterTool; all of them share onedispatchToolcarrying the shutdown gate, in-flight tracking and result formatting. WiresToolRegistry+ProjectManagementServicethroughsrc/container.ts(tsyringe DI).- Tool errors: an execution failure returns
{content, isError: true}so the model can see and recover from it. A JSON-RPC error is reserved for protocol faults (unknown tool). Note v2 validates arguments against the declaredinputSchemabefore the handler runs, soToolValidatornever sees a missing required field. src/domain/— types, zod schemas, errors. No logic.src/services/— business logic.ProjectManagementServiceorchestrates per-feature services;services/ai/holds AI SDK providers.src/infrastructure/—github/(Octokit GraphQL/REST),tools/(MCP tool defs + registry + validators), plus cache, persistence, resilience, events.src/env.ts+src/cli.ts— config. CLI flags override env vars.
Environment (.env or CLI flags; CLI wins)
- Required:
GITHUB_TOKEN,GITHUB_OWNER,GITHUB_REPO. - GitHub credential chain, first hit wins:
--tokenCLI flag →$SECRETS_DIR/GITHUB_TOKENfile →GITHUB_TOKENenv →gh auth token. Theghfallback means a developer with a workingghlogin needs no configuration at all; disable it withGH_CLI_TOKEN_FALLBACK=false(it is also off automatically underNODE_ENV=test, so tests never shell out). - GitHub App installation auth (optional, outranks the PAT when all three are
set):
GITHUB_APP_ID,GITHUB_APP_PRIVATE_KEY,GITHUB_APP_INSTALLATION_ID. A partial App config falls back to the PAT rather than failing, so a half-set environment cannot lock the server out. - Every credential read goes through
getSecret()/requireToken(). Do NOT readprocess.env.GITHUB_TOKENdirectly — that bypasses the whole chain, which is exactly howSECRETS_DIRand--tokenwere silently broken before. - AI (optional, unlocks AI tools):
ANTHROPIC_API_KEY,GOOGLE_API_KEY,OPENAI_API_KEY,PERPLEXITY_API_KEY; model overridesAI_MAIN_MODEL,AI_PRD_MODEL,AI_RESEARCH_MODEL,AI_FALLBACK_MODEL. - Optional:
SYNC_ENABLED,SYNC_TIMEOUT_MS,CACHE_DIRECTORY,WEBHOOK_SECRET,WEBHOOK_PORT,SSE_ENABLED. - Webhook security: signature validation fails closed. With no
WEBHOOK_SECRET, webhooks are rejected unlessWEBHOOK_ALLOW_UNSIGNED=true(trusted dev only). - Secrets: set
SECRETS_DIR(e.g./run/secrets) to load any config/secret from a file named after it (Docker/k8s secret convention), checked before env vars.getSecret(name)inenv.tsreads fresh (rotation-aware). Vault/AWS SM are an extension point viaSecretProvider(src/infrastructure/secrets/). - Startup validation runs against the resolved config, not raw
process.env— validating the environment directly meant a token supplied by file or CLI flag failed the required-field check and the process exited before the resolver ran. - Logs are redacted:
redactSecrets()ininfrastructure/loggerstrips token/secret/password/apiKey/authorization-shaped keys from anything the logger stringifies, andGitHubConfig.tokenis non-enumerable soJSON.stringifycannot emit the PAT.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 238 lines · 3,572 tokens per session scan A a8ab01866304
mcp-github-project-manager AGENTS.md is an instructions file published in the GitHub repository kunwarVivek/mcp-github-project-manager (95 stars, last pushed 21d ago), licensed MIT. It adds 3,572 tokens to every session, about $0.0179 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
pm-skills CLAUDE.md
Instructions for product-on-purpose/pm-skills, covering claude code instructions, style rules, no em-dash characters, documentation rules and public vs private files.
core copilot-instructions.md
Instructions for asdecided/core: These decisions are already accepted. Do not re-open or contradict them; ask the AsDecided MCP tools (getartifact, searchartifacts) for the full text before proposing a change that touches one.
productskills CLAUDE.md
Instructions for assimovt/productskills, covering productskills, what this repo is, architecture, skill design rules and skill.md format.
myscrollr AGENTS.md
AGENTS.md instructions for doughknee/myscrollr, covering agents.md, project overview, repository layout, running it locally and build, lint, test commands.
core CLAUDE.md
Instructions for asdecided/core, covering asdecided — agent session context, loaded every session, working corpus and settled decisions (asdecided).
claude-plugin-product-management CLAUDE.md
Instructions for pmprompt/claude-plugin-product-management, covering claude.md, goals, repo structure, style guidelines for skills and safety.