cc-memory CLAUDE.md

Project instructions for cc-memory, a Claude Code plugin that saves and searches useful information between sessions. It uses files, full-text search, and optional AI-based extraction and deduplication.

In plain words
What is it for?
Use them when changing memory storage, search, plan handoffs, directive tracking, transcript limits, consolidation, or the optional local Ollama fallback.
Why use it?
They document how the plugin prevents repeated or conflicting notes, carries plans between sessions, and observes possible prompt-injection content.

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/skymanbp/cc-memory/claude-md
Clone the repo
git clone --depth 1 https://github.com/skymanbp/cc-memory
Per session 28,206 This file is loaded in full into every session.
When invoked 28,206 The same file — it is already loaded in full.
Security scan D 3 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 $0.28206 $0.28206
Opus 5 $0.14103 $0.14103
Sonnet 5 $0.05641 $0.05641
Haiku 4.5 $0.02821 $0.02821

Measured 2d ago against content hash c6da794e78df, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade D, and why

cc-memory CLAUDE.md scanned grade D with 3 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.

Reads agent configuration directoriesmediumAgent snooping

.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.

`~/.claude/settings.json` → the plugin manifest → `hooks/hooks.json`; the

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

Bash patterns (`git push`, `rm -rf`, deploys) flag drift.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

whitelist — beyond the obvious `sqlite3, json, pathlib, urllib, datetime,
CLAUDE.md · 1,776 lines

How it starts

The opening of the file, as written. The whole thing — 1,776 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md — Project Instructions for Claude Code

Project: cc-memory

Claude Code persistent memory plugin (v2.13.2) — anti-patch reconcile-on-write

  • LLM-judged semantic de-duplication with backpressure-triggered consolidation, forced PROGRESS.md handoff with per-session annotation, live PLAN.md anchor with plan-refiner / plan-guardian subagents + mandatory carryover gate, an enforced directive ledger, bounded transcript reads, injection observability, FTS5 search, AI-judged extraction with Haiku (optional local Ollama fallback).
  • Language: Python 3.8+ (pure stdlib, zero pip dependencies at runtime)
  • Version: 2.13.2
  • License: MIT
  • Platform: Windows-primary, cross-platform compatible (Tkinter required for GUI)

What changed in v2.13.2 (over v2.13.1)

A rename is not finished when the joins are. v2.13.0 swept path joins and tracked markdown; 95 lines of prose inside .py files still spelled memory/<something>, and one of them was not prose: llm/memory_writer._render_memory_index built MEMORY.md's archive links from a hard-coded "memory/" instead of the directory it was handed, so every generated index pointed at a path that had stopped existing. Three rules:

  1. A rename sweep must cover strings the user or the model READS, not only paths the code JOINS. Nine such lines survived v2.13.0: two argparse help= strings, four print()s, and three that core/plan.py renders into PLAN.md — the live plan anchor, telling Claude to open memory/.plan_raw.md. Grep for the old name in help=, print(, and any list of strings a renderer joins, not just for / "name".

  2. A test that spells the same literal as the code cannot catch the code. smoke_test.py § v2.8.0 a6 filtered on "- \memory/sessions/" and so passed on the wrong output for a whole release. Assertions about a rendered path take the name from the fixture (_MEM/_OLDMEM`), never from a literal — and the legacy case gets its own render, because a renderer must name the directory it was given.

Read the full file on GitHub · 1,776 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 · 1,776 lines · 28,206 tokens per session scan D c6da794e78df

Subscribe to this mod's changes

cc-memory CLAUDE.md is an instructions file published in the GitHub repository skymanbp/cc-memory (6 stars, last pushed 2d ago), licensed MIT. It adds 28,206 tokens to every session, about $0.1410 per session on Opus 5. A static security scan graded it D with 3 findings (reads agent configuration directories, recursive force delete, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other instructions, from other repositories