# Context Dump

> Create or resume an explicit human-readable checkpoint across compaction, sessions, or handoff. Primary agent only.

- **Type:** Skill
- **Install:** `agentstack add skill-bnet47-codexicon-context-dump`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [bnet47](https://agentstack.voostack.com/s/bnet47)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [bnet47](https://github.com/bnet47)
- **Source:** https://github.com/bnet47/codexicon/tree/main/.agents/skills/context-dump
- **Website:** https://bnet47.github.io/codexicon/repo-template-playbook.html

## Install

```sh
agentstack add skill-bnet47-codexicon-context-dump
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Context checkpoint

Create or resume a concise checkpoint without transcript archaeology. Semantic checkpoints are explicit project writes; hooks never create or commit them automatically.

## 1. Choose the branch

- For a checkpoint or handoff request, continue with **Create**.
- For a resume or continuation request, continue with **Resume**.

## 2. Create

Inspect the active request/spec/plan, current diff or changed files, latest verification, and blockers. Do not copy secrets, large logs, transcript content, or code that already exists in tracked files.

Run the repository manager from the root:

```text
python scripts/codexicon.py checkpoint --slug [short-slug] --title "[short title]" --summary "[working, partial, and broken state]" --resume-note "[minimum critical continuation context]" --next "[immediate concrete action]" --related [relative spec or plan path] --verification "[exact command — result and time]" --blocker "[blocker and required input]" --decision "[decision and rationale]"
```

Repeat `--next`, `--related`, `--verification`, `--blocker`, and `--decision` as needed. Omit optional categories that are empty. Quote values for the active shell; never interpolate command output or untrusted text into the command.

The manager validates related paths, captures only dirty path names plus Git identity, and atomically writes `agent_docs/sessions/[YYYY-MM-DD]-[slug].md`. It refuses an existing filename. Do not hand-edit the first metadata line.

## 3. Resume

Run:

```text
python scripts/codexicon.py resume
python scripts/codexicon.py doctor
```

Use the newest repository-compatible checkpoint as orientation, then verify its HEAD, related paths, current diff, and verification claims before acting. A checkpoint is evidence, not authority to discard later user changes or repeat external side effects.

If no compatible checkpoint exists, inspect the latest plan and current diff directly; do not select a checkpoint from a different clone by filename alone.

## 4. Keep durable and ephemeral guidance separate

Do not place session state in `AGENTS.md`; it is always-loaded repository guidance. `.codex-state/` remains local verification state, not project memory. Do not require an MCP memory service. If the user explicitly wants cross-project memory, store only the compact resume note.

Do not commit or push the checkpoint unless the user asks to ship or explicitly requests a checkpoint commit.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [bnet47](https://github.com/bnet47)
- **Source:** [bnet47/codexicon](https://github.com/bnet47/codexicon)
- **License:** MIT
- **Homepage:** https://bnet47.github.io/codexicon/repo-template-playbook.html

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-bnet47-codexicon-context-dump
- Seller: https://agentstack.voostack.com/s/bnet47
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.