agentsclimarketplace

Local llm sessions

Skill qiaoruntao/skills/local-llm-sessions

Public Claude Code skills

Install
npx -y skills add qiaoruntao/skills --skill local-llm-sessions

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Read, list, search, and fetch local AI coding assistant conversation history from Claude Code, Codex CLI, Gemini CLI, OpenCode, Antigravity, Cline, Cursor, Aider, and ForgeCode. Use when Codex needs to inspect local LLM sessions, compare provider history, recover prior conversation content, list recent sessions, fetch message transcripts, or understand raw local session storage formats.

SKILL.md

3.9 KB, 841 tokens by cl100k_base, as published. Nobody here has run it

Local LLM Sessions

Core Rule

Use the local CCHV server API first. It normalizes multiple providers into the same project/session/message shape and avoids fragile direct parsing of vendor-specific files.

Default local service:

  • URL: http://127.0.0.1:3727
  • Optional PM2 process name: cchv-server
  • Auth token source: CCHV_TOKEN environment variable, or pass --token

If the API is unavailable and the user manages CCHV with PM2, start or inspect it with:

pm2 status cchv-server
pm2 logs cchv-server --lines 100

Do not scan every provider by default. Use provider filters and request timeouts; some providers can be slow on large local trees.

Quick Start

Use the bundled script for routine reads:

python3 skills/public/local-llm-sessions/scripts/cchv_sessions.py providers
python3 skills/public/local-llm-sessions/scripts/cchv_sessions.py projects --provider codex
python3 skills/public/local-llm-sessions/scripts/cchv_sessions.py sessions --provider claude --project-path '<project path>'
python3 skills/public/local-llm-sessions/scripts/cchv_sessions.py messages --provider codex --session-path '<session file path>' --transcript

When installed as a Codex skill, use the installed skill path instead of the repository path above.

Prefer --transcript when the user wants readable conversation content. The default transcript is a human/debug view: it skips developer/system boilerplate, permission blocks, app context, progress/attachment records, and truncates huge tool outputs.

Use these transcript controls when debugging interaction quality:

  • --forensic: preserve boilerplate and full tool output.
  • --no-tools: show only human/assistant text and thinking, hiding tool calls/results.
  • --no-thinking: hide thinking blocks.
  • --max-tool-chars 0: do not truncate tool blocks.

Use raw JSON when they need exact fields, complete tool payloads, or provider-specific metadata.

API Workflow

  1. Resolve auth token.
  2. Check GET /health.
  3. List projects for one provider: POST /api/scan_all_projects with {"activeProviders":["codex"]}.
  4. Pick a project path.
  5. List sessions: POST /api/load_provider_sessions with {"provider":"codex","projectPath":"...","excludeSidechain":false}.
  6. Pick a session file_path.
  7. Fetch messages: POST /api/load_provider_messages with {"provider":"codex","sessionPath":"..."}.

Provider ids observed in CCHV: claude, codex, gemini, antigravity, opencode, cline, cursor, aider, forgecode.

Message Handling

Normalized messages commonly include:

  • provider
  • type
  • role
  • timestamp
  • uuid
  • sessionId
  • parentUuid
  • content
  • toolUse
  • toolUseResult
  • provider-specific extras such as data, usage, or toolUseID

When building transcripts:

  • Preserve ordering returned by the API.
  • Include timestamps and provider when useful.
  • Treat content as JSON-like: it may be a string, object, list of text blocks, or null.
  • Default to the human/debug transcript for understanding how the user and model interacted.
  • Filter environment boilerplate unless the user asks for forensic reconstruction.
  • Include tool calls/results only when the user asks for tool activity or when they explain the answer; otherwise consider --no-tools.
  • Avoid dumping huge sessions verbatim; summarize or page unless the user asks for full content.

Raw Format Reference

Read references/session-formats.md when direct file inspection is necessary, the CCHV API is unavailable, or you need provider-specific storage details.

What ships with it: 3 files

13.6 KB alongside SKILL.md, 1 of them executable

agents/

references/

scripts/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.