code-recallSAFE
Local decision ledger for AI coding agents — keeps why a choice was made and which paths are dead ends, and re-injects it after context compaction. Plain Markdown, zero dependencies, works across Claude Code / Codex / Cursor / Copilot / Gemini.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
English · 繁體中文
Long session, the context window compacts — and your agent re-opens a decision you settled an hour ago, re-walks a dead end you already ruled out, sometimes rewrites code it just wrote the opposite way. The chat had the reasoning; compaction ate it. Code Recall keeps it in front of the agent at every reset — instead of hoping the model recalls it.
[](https://opensource.org/licenses/MIT) [](https://github.com/erikhuang76821/code-recall/actions/workflows/ci.yml) [](https://www.npmjs.com/package/@erikhuang/coderecall) [](http://makeapullrequest.com)
Code Recall is a tiny local decision ledger for AI coding agents. It holds what projects lose most easily and rebuild most expensively: why a choice was made, and which paths are proven dead ends.
Three plain Markdown files in your repo. A SessionStart hook puts them back in front of the agent the moment context resets — session start, resume, and after compaction — and a PreCompact hook snapshots the conversation tail just before it is destroyed. (Native hooks on Claude Code; instruction-driven on other tools, and verified injecting into Codex CLI unchanged.)
Not a memory database, not the cloud, not a governance platform. Zero dependencies, travels with the repo. Think "Git for decisions" that survives compaction.
Requirements · Node ≥ 10.12 (anythin
12c9848b6ca0OBSERVED · 2026-10-12Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.
claude mcp add coderecall -- npx -y @erikhuang/[email protected]
{
"mcpServers": {
"coderecall": {
"command": "npx",
"args": [
"-y",
"@erikhuang/[email protected]"
]
}
}
}Exposed tools (7)
3 read · 4 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
list_decisions | write | List CURRENT (accepted) decisions — the HEAD view. Set |
read_memory | read | Read the project memory digest (GOAL/NOW/NEXT, counts, blocked items). |
reconfirm | read | Re-stamp an entry as still-true today without rewriting it: refreshes |
resolve_lesson | read | Mark a LESSONS entry no longer current: |
update_task | write | Update TASK.md GOAL/NOW/NEXT. Provide any subset. |
write_decision | write | Record an ADR-grade decision (DECISIONS.md): why + what + consequences. Provide either |
write_lesson | write | Record a lesson / failure + root cause (LESSONS.md): |
Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | PASS |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (5 observation(s))
- Network
- declared (1 observation(s))
- Shell
- declared (4 observation(s))
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (2)
docs/assets/1.png
docs/assets/2.png
Gates applied: no_behavioural_pass.
12c9848b6ca0full audit observations/trust-audit/mcp-server/erikhuang76821__code-recall.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-12 | 12c9848b6ca0 | SAFE | B | 89 | first audit |
Questions
What is the code-recall MCP server?
Local decision ledger for AI coding agents — keeps why a choice was made and which paths are dead ends, and re-injects it after context compaction. Plain Markdown, zero dependencies, works across Claude Code / Codex / Cursor / Copilot / Gemini.
What tools does code-recall expose?
7 in total: 3 read-only, 4 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is code-recall safe to connect to an agent?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean server reads B.
What credentials does code-recall need?
No credential environment variables were found in its source, so it appears to need none.
How current is this page?
The grade is for one exact copy of the source (12c9848b6ca0), read on 2026-10-12. The repository is watched and re-audited when it changes.