Pachigraph plugin

This package provides a portable Agent Skill with Python HTTP tools. It supports the Agent Plugins 1.0 and Codex layouts, and can also be installed as a standalone skill. No MCP server registration is required.

Generate a read key from the signed-in /keys page.

Install the skill

Copy skills/pachigraph/ into your agent's skill directory. For Codex, use ~/.codex/skills/pachigraph/ (or $CODEX_HOME/skills/pachigraph/). Keep its SKILL.md and scripts/query.py together. Invoke $pachigraph in a task and provide the key-file path, never the key value. The agent must have a shell tool and permission to reach the configured HTTPS API.

Clients supporting Agent Plugins can load this package's plugin.json and skills/; Codex plugin clients can use .codex-plugin/plugin.json. Neither manifest contains credentials or a remote tool-server configuration.

HTTP CLI

Save the key as one UTF-8 line in a file owned by the current user with no group or world permissions, then run:

chmod 600 /absolute/path/to/key
python3 skills/pachigraph/scripts/query.py \
  --token-file /absolute/path/to/key search 'query terms'
python3 skills/pachigraph/scripts/query.py \
  --token-file /absolute/path/to/key fetch RECORD_ID
python3 skills/pachigraph/scripts/query.py \
  --token-file /absolute/path/to/key status

The CLI uses only Python's standard library, rejects redirects, and accepts plain HTTP only for loopback testing.

The commands call /api/search, /api/fetch, and /api/status with the read key. status.text_bytes measures indexed text only, not allocated D1 storage. The collector retains its separate ingest key. Live HTTP gateway qualification is still required; installing the skill does not change Sites access controls.