agentdir

A repository rule that uses AgentDir, a background record of agent work, to capture evidence while engineers work in an unfamiliar codebase.

In plain words
What is it for?
Use it when starting engineering work to check the repository setup, begin a work record, read related context, and close the review after the task.
Why use it?
It helps agents find relevant prior work and preserve a searchable record of the decisions and changes made during development.

Cursor rule for Cursor

Install

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.

agentmods
npx agentmods add rules/jstxn/agentdir/agentdir
Clone the repo
git clone --depth 1 https://github.com/jstxn/agentdir

Made for: Cursor.

Per session 631 This file is loaded in full into every session.
When invoked 631 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00631 $0.00631
Opus 5 $0.00316 $0.00316
Sonnet 5 $0.00126 $0.00126
Haiku 4.5 $0.00063 $0.00063

Measured 2d ago against content hash 8b885e96c7f9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agentdir 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 2d 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.

.cursor/rules/agentdir.mdc · 26 lines

What it actually says

AgentDir for Cursor

AgentDir is agent-owned background instrumentation for this repository. The engineer installs or adopts it once; agents operate it during normal work.

  • Before starting work in an unfamiliar checkout, run agentdir root --require --quiet. Exit 0 means an initialized local or shared worktree store is ready; do not decide whether AgentDir is set up from the presence of .agentdir in this checkout.
  • If the probe exits 3, run agentdir adopt --if-needed --gitignore user. If hook installation is blocked in a restricted linked worktree, rerun it with --no-hooks.
  • Then start non-trivial coding work yourself with agentdir work start "<subsystem: distinctive behavior or constraint>". Keep the title concise but retrieval-specific: name the subsystem plus a distinctive behavior or constraint, preserving request terms likely to match prior work.
  • Read the printed context briefing, treat excerpts as previews, and prefer the printed agentdir work context --pack <pack-id> --expand <number> command before using a source when implementation details, prior patterns, or exact evidence matter. Then close the review once: repeat --use in that command for every useful numbered source (agentdir work context --use <number> [--use <number> ...] --reason "<how they help>"), or run agentdir work context --none-relevant --reason "<why>". Use --skip --reason "<why>" only when review is impossible; no decision is needed when no sources are presented. Re-open lost output with agentdir work context --show.
  • Wrap evidence-bearing commands with agentdir run -- <command>.
  • Evidence includes tests, lint, typecheck, builds, doctor checks, release checks, reproduced failures, and diagnostics used in final claims.
  • Do not wrap routine exploration such as rg, sed, nl, cat, ls, find, or quick read-only git status.
  • Use agentdir evidence --brief and agentdir timeline to skim what happened.
  • Record each verification you will rely on with agentdir claim <family> --passed|--failed, e.g. agentdir claim test --passed. Families: test, lint, typecheck, build, doctor, release.
  • Claim what the evidence shows, including failures. agentdir audit claims compares recorded claims against evidence, so a claim that overstates a result is reported as contradicted.
  • Re-recording a family replaces its earlier claim; withdraw one made in error with agentdir claim <family> --retract.
  • Use agentdir report final --format json or agentdir work finish --json --brief for the agent handoff object before final claims when practical.
  • Do not record secrets, private keys, raw environment dumps, or credential-bearing command output.
Changes

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.

  1. 2d ago First seen · 26 lines · 631 tokens per session scan A 8b885e96c7f9

Subscribe to this mod's changes

agentdir is a cursor rule published in the GitHub repository jstxn/agentdir (21 stars, last pushed 11d ago), licensed MIT. It adds 631 tokens to every session, about $0.0032 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.