Skill

workspace

Initializes, checks, and refreshes Retriever workspaces: sets up .retriever directories, migrates schema, verifies runtime via workspace status/init/update, and manages tools.py bundle.

Python

SQLite

developer-tools

automation

npx claudepluginhub sdemyanov/retriever

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Use this skill for any task that needs to bootstrap or maintain a workspace-local Retriever installation.

Supporting Assets

requirements.lock.mdworkspace.md

SKILL.md

Similar Skills

ingest

Ingests, indexes, imports, loads, or refreshes files and folders into Retriever workspaces. Bootstraps if needed, chooses ingest mode, and summarizes results.

retriever

init-workspace-context

Classifies workspace initialization mode (install, upgrade, plugin) and builds inventory of existing Rosetta bootstrap files. Detects composite multi-repo setups.

core

openclaw-update

586

Backs up, updates, and restores workspaces with dynamic detection using git, tar, jq, and bash. Ideal for safe Claude Code environment management.

sundial-org-awesome-openclaw-skills-4

Stats

Stars10

Forks0

Last CommitApr 28, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

Retriever Workspace

Use this skill for any task that needs to bootstrap or maintain a workspace-local Retriever installation.

Load order

Read workspace.md first.
Read requirements.lock.md when dependency installation or verification is involved.
Read ../tool-template/tool-template.md when using or refreshing the canonical tools.py bundle.
Read ../schema/schema.md when initializing or migrating the database schema.

Core rules

The workspace root is the user-selected folder, not .retriever/.
All persistent Retriever state lives under .retriever/ inside that root.
Database paths stored in SQLite must be relative, never absolute.
Treat skills/tool-template/tools.py as the canonical command entrypoint.
Use workspace status, workspace init, and workspace update; old direct doctor, bootstrap, and upgrade-workspace command names are implementation/history details, not the user-facing surface.
The tool's own dispatcher refreshes runtime metadata for stale workspaces before running any non-exempt command; see workspace.md for the full contract.
Treat runtime.json as the local state record for the canonical tool version, schema version, and checksum.
Treat canonical template checksum drift as an upgrade signal even if the plugin version string is unchanged.
On reinstall, the next non-exempt command will refresh runtime metadata automatically; there is no separate reindex step needed purely because the canonical template changed.
If environment checks fail, stop and report the issue clearly instead of partially bootstrapping.

Current outcome

With the current 1.1.11 / schema 25 tool surface, Claude should be able to:

verify the runtime with workspace status
create the .retriever/ directory structure
use the canonical tools.py bundle directly
initialize or migrate schema 25
write a stable runtime.json
ingest supported documents through the bounded V2 ingest facade
verify the required PST backend is present before calling the runtime ready
populate built-in metadata for email, chat-like, PST, MBOX, Slack, and production documents during ingest
ingest supported PST container sources through the same regular ingest surface
materialize one-level EML/MSG attachment families with stable control_number values during ingest
search indexed documents with structured filters

Do not improvise the runtime contract beyond the referenced files.