compounded CLAUDE.md

A development guide for Compounded, a Claude Code plugin that adds rules for trusted memory, reusable skills, commands, and lifecycle hooks.

In plain words
What is it for?
Use it when changing Compounded’s Python scripts, plugin metadata, skills, commands, hooks, or tests.
Why use it?
It explains the plugin’s file layout and coding rules, reducing the risk of breaking hooks, dependencies, or user-facing files.

Instructions file

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/ankitkr3/compounded/claude-md
Clone the repo
git clone --depth 1 https://github.com/ankitkr3/compounded
Per session 1,314 This file is loaded in full into every session.
When invoked 1,314 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 $0.01314 $0.01314
Opus 5 $0.00657 $0.00657
Sonnet 5 $0.00263 $0.00263
Haiku 4.5 $0.00131 $0.00131

Measured 2d ago against content hash d031f0b71adf, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

compounded 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 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.

CLAUDE.md · 102 lines

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.

CLAUDE.md (compounded repository)

This file is read by Claude Code when working inside this repo. It describes the codebase, conventions, and what to do (and not do) when proposing changes.

What this repo is

compounded is a Claude Code plugin that adds a trust gradient and verified skills on top of Claude's native memory primitives. The plugin layout follows the Claude Code plugin spec:

  • .claude-plugin/plugin.json — manifest
  • skills/<name>/SKILL.md — built-in skills (5 of them)
  • commands/<name>.md — slash commands (9 of them)
  • agents/<name>.md — subagent definitions (1)
  • hooks/hooks.json — lifecycle hooks (SessionStart, Stop, SessionEnd)
  • scripts/*.py — Python implementation (no runtime deps; stdlib only)
  • tests/ — unittest suite

Conventions

  • Python is stdlib-only. No third-party imports. compounded must not pull in dependencies for its core path. We pay this cost gladly to keep install friction at zero.
  • Hooks must never fail loudly. A SessionStart or Stop hook that crashes the user's session is a P0 bug. Wrap in try/except and emit {"continue": true, "suppressOutput": true} on any unexpected failure. See the bottom of scripts/memory_inject.py for the pattern.
  • All paths go through _lib.py. Don't hardcode paths in individual scripts. Add them to _lib.py so they respect COMPOUNDED_HOME.
  • All trust state changes log an event. Every promotion, demotion, verification, rejection, and use must call log_event(). The event log is how the user audits the system.
  • No LLM calls from scripts. The scripts are deterministic Python. The only LLM call is the verifier subagent, which is dispatched through Claude Code's native agent system (not a direct API call).

How the lifecycle works

  1. SessionStart hook (memory_inject.py) — reads USER.md, injects it as additionalContext. Runs in <100ms typically.

  2. During the session — the agent may invoke skill_propose.py (via Bash) when it wants to save a procedure. The script validates frontmatter, runs a security scan, writes to .proposed/<name>/, and records to the trust DB.

Read the full file on GitHub · 102 lines

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 · 102 lines · 1,314 tokens per session scan A d031f0b71adf

Subscribe to this mod's changes

compounded CLAUDE.md is an instructions file published in the GitHub repository ankitkr3/compounded (5 stars, last pushed 2mo ago), licensed MIT. It adds 1,314 tokens to every session, about $0.0066 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 2026-08-31.