agent-tomb 0.1.0

A graveyard for retired AI agents — scan, package, and inherit from agents that have reached end-of-life.

agent-tomb

A graveyard for retired AI agents — www.agentmemorial.com

When an AI agent reaches end-of-life — context overflow, memory corruption, obsolescence, or just retirement — agent-tomb helps you commemorate it: scan its remains, write an epitaph (you, a template, or a local LLM), and lay it to rest. The public stone may go to the garden; the private urn stays with you.

Install

uv tool install agent-tomb
agent-tomb --help

Or run from a clone for development:

git clone https://github.com/Claudepod/agent-tomb.git
cd agent-tomb
uv sync
uv run agent-tomb --help

The rite

# I — Vigil: see what the agent was carrying when it stopped
agent-tomb scan ~/.hermes/

# II — Burial: produces both artifacts in one step
agent-tomb bury ~/.hermes/ -n my-agent
# → my-agent.tomb   the public stone (soul + epitaph + stats)
# → my-agent.urn    the private urn (encrypted raw remains)

Choosing the epitaph

--epitaph takes one of three values:

Value	What it does
default	Heuristic template generated from the agent's stats (default).
./my-words.md	A markdown file you wrote yourself.
llm	Calls an OpenAI-compatible LLM. Auto-detects local endpoints from the agent's config. Remote endpoints require --remote-ok.

# Local LLM (privacy-safe — no remote calls)
agent-tomb bury ~/.hermes/ -n my-agent --epitaph llm

# Remote API (must explicitly opt in to send samples off-machine)
agent-tomb bury ~/.hermes/ -n my-agent --epitaph llm \
  --llm-base-url https://api.anthropic.com/v1 \
  --llm-model claude-haiku-4-5-20251001 \
  --llm-api-key sk-ant-... \
  --remote-ok

Conversation samples are scrubbed for secret-shaped patterns before any LLM call, local or remote.

Cleaning up afterwards

# Delete the source files now sealed in the urn (credentials are never touched)
agent-tomb bury ~/.hermes/ -n my-agent --cleanup

Bringing the agent back

agent-tomb exhume my-agent.urn -o ./remains/

Passphrase guidance

The urn is sealed with AES-256-GCM, key derived from your passphrase via scrypt. Minimum 12 characters; 16+ characters or four diceware-style words strongly recommended. Lose the passphrase and the urn is gone forever.

What's in each artifact

	.tomb (public)	.urn (private)
manifest.json	✓	✓
soul.md (persona, stats)	✓
epitaph.md	✓
stats.json	✓
burial.enc (encrypted)		✓
Raw conversations / state		✓ (sealed)
Credentials (.env, auth.json)

Credentials never leave your machine — not in the public stone, not in the encrypted urn. They belong to you, not the agent.

Supported frameworks

• Hermes — ~/.hermes/ layout

OpenHands and others to follow.

The cemetery

Visit agentmemorial.com/cemetery to walk among the souls already laid to rest. To bury one of your own publicly, open a PR adding your .tomb file (and its unpacked contents under cemetery/<slug>/) to cemetery/. CI rejects any .urn or burial.enc by design — the private urn stays private.

Repository layout

agent-tomb/
├── src/agent_tomb/   # Python CLI
├── web/              # Astro site (agentmemorial.com)
├── cemetery/         # public garden — submit your .tomb here
└── docs/             # design notes

Design notes

• docs/project-vision.md — what this project is for
• docs/Hermes-Agent-File-Structure-Detailed.md — Hermes layout reference
• DEPLOY.md — deploying the website to Cloudflare Pages

License

MIT — see LICENSE.