agenttrail AGENTS.md

Repository instructions for AgentTrail that define how to maintain its living PLAN.md file. The plan records concrete work outcomes, task status, ownership, and technical notes for the project owner.

In plain words
What is it for?
Use them when updating AgentTrail’s plan, assigning work, recording implementation details, and maintaining its component and task map.
Why use it?
They keep progress visible and make it clear which work is pending, underway, finished, or blocked.

Instructions file for CodexOpenCode

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 instructions/sodiumsun/agenttrail/agents-md
Clone the repo
git clone --depth 1 https://github.com/sodiumsun/agenttrail

Made for: Codex, OpenCode.

Per session 674 This file is loaded in full into every session.
When invoked 674 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 $0.00674 $0.00674
Opus 5 $0.00337 $0.00337
Sonnet 5 $0.00135 $0.00135
Haiku 4.5 $0.00067 $0.00067

Measured 2d ago against content hash 2b0d80272ccf, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agenttrail 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 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.

Origin

Copies of this mod

1 near-identical copy found in the catalogue:

AGENTS.md · 16 lines

What it actually says

agenttrail plan convention

Maintain PLAN.md as the living plan. It is read by the project OWNER, not by you — write it for them.

  • nodes are COMPONENTS of the system being built (## Plain-language name {#id}), not phases or sprints; keep the map at 5-9 components regardless of repo size — grow tasks, not cards, and split a component only when one agent could no longer own it for a session
  • naming rule: titles are verb-led, plain-language, and CONCRETE — the owner can tell when it is done ("Read alerts out loud", "Watch the repo"). Never engineer-speak ("fs watcher + activity signal") and never vague vibes ("Decide what matters"); put the engineer phrasing on a tech: line under the heading
  • tasks inside a component: - [ ] Plain outcome {#id}, optional indented tech: line beneath; mark a task [~] BEFORE you start it and save PLAN.md immediately — this drives the live in-progress view; flip it to [x] the moment it completes, [!] if stuck (clear once unblocked). Never batch plan updates for the end of the session
  • when you mark a task [~], add an indented by: <your name> line under it (claude, codex, cursor, …) and leave it there when done — it is the record of who did what
  • edges under a component heading: needs: [id, id] = must come after those components; links: [id, id] = interconnected with / talks to
  • files: [src/audio/**, config.py] under a component declares which paths it owns — keep it current; it is how the live view knows which component you are really working in, including when you revisit finished work
  • {#id}s are stable — never rename, only add or remove nodes
  • open tasks carry an indented from: line naming their provenance — from: agent when YOU are declaring it as your own imminent build intent (the owner corrects these on sight if wrong), from: roadmap when it comes from planning documents (durable intent, backloggable); omit when neither
  • new work NEVER creates a component by default: it lands as your session todos plus tasks under the component whose files it touches. Add a NEW component only when the system grows a durable new part. Durable is not a prediction — test it NOW: something else already depends on it (load-bearing), deleting it would change what the product does, you can name a plausible second task for it, and it owns files no component claims. Record the addition under ## decisions. Time corrects mistakes cheaply: merge a card back into its neighbor if no second change ever comes; promote a task to a card when work keeps clustering in files its siblings never touch. Remove a component only when that part is deleted from the product
  • before ending a session, graduate your plan-worthy completed todos into PLAN.md as [x] tasks (with by:) — housekeeping todos stay out of the plan
  • record any plan-affecting decision under ## decisions BEFORE implementing it
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 · 16 lines · 674 tokens per session scan A 2b0d80272ccf

Subscribe to this mod's changes

agenttrail AGENTS.md is an instructions file published in the GitHub repository sodiumsun/agenttrail (474 stars, last pushed 3d ago), licensed MIT. It adds 674 tokens to every session, about $0.0034 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.