Borrowing it
Nothing to install: this file belongs to keshrath/agent-tasks. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/keshrath/agent-tasks/main/CLAUDE.mdgit clone --depth 1 https://github.com/keshrath/agent-tasksWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/keshrath/agent-tasks/claude-md)<a href="https://agentmods.dev/instructions/keshrath/agent-tasks/claude-md"><img src="https://agentmods.dev/badge/instructions/keshrath/agent-tasks/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/keshrath/agent-tasks/claude-md"><img src="https://agentmods.dev/badge/instructions/keshrath/agent-tasks/claude-md.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.02695 | $0.02695 |
| Opus 5.5 | $0.01078 | $0.01078 |
| Sonnet 5.5 | $0.00539 | $0.00539 |
| Haiku 4.5 | $0.00269 | $0.00269 |
Grade A, and why
agent-tasks CLAUDE.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 4d ago.
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.
How it starts
The opening of the file, as written. The whole thing — 102 lines — stays where its author put it; the contents beside it link to each section on GitHub.
agent-tasks
Architecture
Layered architecture with explicit dependency injection (no global state):
src/
index.ts MCP stdio entry point; one process serves the REST API (leader election on the port; the others retry and take over)
server.ts REST server (standalone: node dist/server.js [--port] [--db])
domain/ tasks (stages, dependencies, artifacts, claiming), events, validation
storage/ SQLite (better-sqlite3, WAL mode)
transport/ REST (node:http, JSON only) + guard.ts (Host/Origin/Content-Type), MCP (official SDK `Server` over stdio, mcp.ts)
plugin/ Claude Code plugin (.mcp.json + one mod: hooks/register.tsx + view.tsx = /pipeline pane)
- No frameworks — no React, Vue, Express. Pure Node.js + TypeScript; runtime deps are the official MCP SDK (
@modelcontextprotocol/server) and better-sqlite3. No agent-common. context.tsis the DI root — wires all services together.src/stays host-agnostic; Claude-specific glue belongs inplugin/(the mod) orscripts/hooks/(command hooks wired byscripts/setup.js).- Soft deps:
AgentBridge→ agent-comm (notifications),KnowledgeBridge→ agent-knowledge (push learnings/decisions on task completion). Both HTTP-only, fail-open.
UI surfaces
There is no web dashboard (removed in 2.0, with the WebSocket, the UI files and the agent-desk plugin).
- Claude Code pane (
plugin/hooks/register.tsx+view.tsx, Claude Code 2.1.289+ function hooks):/pipelineopens it (the built-in/tasksis taken);/pipeline 12opens task 12,/pipeline <text>filters the board. Tabs Board (open tasks by stage, Find over id, title, agent, project and tags,waits for #Non blocked rows), Done (recently closed, who and where, the result beneath), New (title, project, priority, tags, description; Enter walks the fields). Task detail: stage strip, actions for a task in progress (Advance › next, Complete, ‹ Send back with a stage, Fail; Cancel, Delete with a confirmation), facts, description, subtasks, artifacts (unfold as Markdown), comments + a comment field. No Claim: agents claim withtask_stage, and an assignee that is no live agent-comm agent gets reaped by the cleanup; an unclaimed task has no Advance either (it would sit in progress with nobody on it). A blocked task shows no Advance or Complete (the domain refuses both while ablocksdependency is open), only what it waits for. A stage gate is said under the actions fromtask_get'sgate_needs(TaskService.gateNeeds, the gate's one reader;checkStageGaterefuses Advance and Complete on it), said with who gives it (gateHint); Advance and Complete loseprimarywhile it holds. The ring never lands in a text field when/pipelineopens or moves to a view (a later/pipeline Nwould be typed into it and posted): a comment gate sends it to the comment field only after a press the gate refused (toCommentIfGated, which also shortens that refusal to✗ #N can't leave <stage> yetsince the hint says why, and keeps a refused Complete's form open with its words and the hint under it); the New tab is the one view that starts in a field. A stage nobody works on folds past 10 rows (boardGroups; board folds survive opening a task); unfolded, its toggle stays after row 10 and the ring goes to the first row it showed. Status line entry and the board's summary are the same text (headline()in view.tsx); toasts only for tasks that finished or reached review since the last poll; oneprompt.contextblock with what is in flight. Acts asAGENT_TASKS_USER, else the OS user (created_by, comment author,by). PollsGET /api/boardevery 5 s while open, 30 s while closed. REST only, through$.http.fetchonAGENT_TASKS_PORT(default 3422). View state in$.state(contractplugin/types/index.d.ts). Opened as a plain sidebar (nocloseOnEscape);/pipelinereportsisPlaced: falsewith the reason. Focus lands explicitly after every navigation and action, a fold's toggle included (landing(),toggle()in register.tsx): the engine keeps the ring at its place in the focus order across a redraw, not on a key, so a redraw that adds or drops a control above it would move it; a key that names the new state (foldKey,advance:<stage>,complete:<stage>fromcompleteKey, the freshcomment:<rev>, Refresh'srefresh:<rev>fromrefreshKey) makes the focus wait for the drawing that shows it;landOnNewBoardwaits on Refresh, a button, so nothing typed meanwhile lands in a field. Back up after the server was away, the ring returns where it was before (landBack), else lands as on a fresh open. The pane's own moves and its autoFocus taking the ring both reachui.focusas originplugin;focusOnmarks its own (ownMoves, a set: moves overlap), so a take (ctrl+x tab) is put back on the lastfocusedkey while it is drawn, andopenPaneclearsfocusedso a command's view starts at its landing. The elementlanding()names is also drawnautoFocus, and/pipelinecloses and reopens an open pane after setting the view, because a take would otherwise put the ring back on the element it last held. Every function that takes$must be a top-level declaration (claude plugin validateenforces it). Tests:plugin/tests/*.test.tsunderclaude plugin test plugin(npm run plugin:check); vitest excludesplugin/. Type-check with a tsconfig outside the folder (nevertsc -p plugin, it emits .js into plugin/). - MCP tool results (JSON + the pipeline footer) for every other host.
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.
- 4d ago Changed · +1 lines · +532 tokens per session 16b9f8aea7f8
- 5d ago Changed · -2 lines · +844 tokens per session 14e0fb1c1b6e
- 1mo ago First seen · 103 lines · 1,319 tokens per session scan A 41057ef18653
agent-tasks CLAUDE.md is an instructions file published in the GitHub repository keshrath/agent-tasks (22 stars, last pushed 5d ago), licensed MIT. It adds 2,695 tokens to every session, about $0.0108 per session on Opus 5.5. 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
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, cache eligibility and rollout and applying the notice (cutover).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
vscode copilot-instructions.md
Copilot instructions for microsoft/vscode, covering vs code copilot instructions, project overview, root folders, core architecture (src/ folder) and built-in extensions (extensions/ folder).
deepseek-harness AGENTS.md
AGENTS.md instructions for deepseek-ai/deepseek-harness, covering agents.md, pre-stable apis and released session data, repository layout, commands and host sandbox failures.