honcho-cli — Claude Code skill: debug Honcho memory
A skill that walks your agent through inspecting Honcho workspaces with the CLI: peer memory, session context, dialectic quality.
npx skills add plastic-labs/honcho --skill honcho-cli -a claude-codeWhat honcho-cli does
This skill hands the agent a working map of honcho, the CLI wrapper around the Honcho Python SDK. Output defaults to JSON when piped, errors come back structured, and input gets validated before it reaches the server.
What's inside: command groups for workspace, peer, session, scope, message and conclusion, plus an outside-in inspection order. Start at the workspace, list peers, read a peer card and its conclusions, then dig into a session transcript. Three debugging playbooks cover the usual complaints — a peer that never learns, a session context that looks off, dialectic answers that miss the point.
The skill limits the agent to Bash(honcho:*), Bash(jq:*), file reads and grep. Credentials and the environment URL live in ~/.honcho/config.json; honcho init sets them. Scope per command goes through -w, -p, -s or HONCHO_* variables. honcho start brings up a local Docker stack but leaves environmentUrl alone, so point at it with HONCHO_BASE_URL=http://127.0.0.1:8000. Deleting a workspace without inspecting it first is off the table.
What it helps with
- find out why a peer stops accumulating conclusions
- see the exact context an agent receives in a session
- search conclusions by topic for one observer
- narrow recall to a project with a scope
- confirm a recall/record loop is actually running
Example prompts for the agent
- Inspect the workspace state with
honcho workspace inspect --json - The peer isn't learning — check its configuration and conclusions
- Show me the session context and recent messages so I can see what the agent gets
How to install honcho-cli
1Skills CLI (Vercel)
npx skills add plastic-labs/honcho --skill honcho-cli -a claude-code2Manually: put the files in place
~/.claude/skills/honcho-cli/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/honcho-cli/
1Skills CLI (Vercel)
npx skills add plastic-labs/honcho --skill honcho-cli -a codex2Manually: put the files in place
.agents/skills/honcho-cli/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add plastic-labs/honcho --skill honcho-cli -a cursor2Manually: put the files in place
.cursor/skills/honcho-cli/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install plastic-labs/honcho honcho-cli2Manually: put the files in place
.github/skills/honcho-cli/SKILL.md1Skills CLI (Vercel)
npx skills add plastic-labs/honcho --skill honcho-cli -a gemini-cli2Platform CLI
gemini skills install https://github.com/plastic-labs/honcho --path skills/honcho-cliManually: put the files in place
.devin/skills/honcho-cli/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add plastic-labs/honcho --skill honcho-cli -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about honcho-cli
How do I install honcho-cli in Claude Code?
Run npx skills add plastic-labs/honcho --skill honcho-cli -a claude-code. The skill lands at ~/.claude/skills/honcho-cli/SKILL.md.
What does honcho-cli do?
It gives the agent commands to inspect a Honcho deployment: peer memory state, session context and transcripts, conclusion search, and direct dialectic chat.
What has to be set up first?
The API key and environment URL in ~/.honcho/config.json, which honcho init writes. Workspace, peer or session scope comes from -w, -p, -s flags or HONCHO_* variables.
