From retriever
Ingests, indexes, imports, or processes files and folders into the Retriever workspace. Routes between ingest and ingest-production, bootstraps the workspace if needed, and summarizes results.
How this skill is triggered — by the user, by Claude, or both
Slash command
/retriever:ingestThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> Operates under `retriever:routing`. If the user's intent actually fits a different tier — another `retriever:*` skill, a Tier 2 slash, a Tier 3 `tools.py` subcommand, or (last resort) direct DB access — stop and re-route against the ladder before continuing.
Operates under
retriever:routing. If the user's intent actually fits a different tier — anotherretriever:*skill, a Tier 2 slash, a Tier 3tools.pysubcommand, or (last resort) direct DB access — stop and re-route against the ladder before continuing.
Use this skill when the user says things like:
python3 skills/tool-template/tools.py workspace status --quick <workspace> before the first ingest. For already-known workspaces, run it whenever runtime state is unclear.workspace init or the first ingest fails with SQLite WAL, journal-mode, mount, or sandbox wording, stop normal ingest execution and follow the mounted/sandboxed SQLite bootstrap troubleshooting.df, mount, or host filesystem labels alone..retriever/retriever.db path.ingest for normal Claude Code indexing and refresh tasks. It is the bounded V2 facade by default.python3 skills/tool-template/tools.py ingest <workspace> --recursiveingest-start / ingest-run-step unless you need to inspect or recover a specific active run.ingest will fail with active_ingest_run; resume or cancel that run intentionally instead of starting another one..pst sources, use regular ingest, not ingest-production.workspace status reports pst_backend.status == fail, explain that PST ingest needs the required libpff-python / pypff backend installed.--path <production-root-relative-path> and no --file-types; it will auto-route the production.ingest-production only when the user explicitly asks for that command, when running a targeted script outside the normal bounded Claude Code command window, or when debugging parity with the production-only path.DAT + companion TEXT/, IMAGES/, and optional NATIVES/ folders as a production signature, not as loose files.ingest with --recursive when the user wants the whole tree scanned.--path <relative-path> flags.ingest without --file-types now auto-routes detected production roots through the production ingest pipeline while still indexing loose files elsewhere in the tree.ingest reports skipped_production_roots, explain that Retriever detected a processed production but skipped it because a file-type filter was present; use ingest-production when the user's intent is to index that production root explicitly.new, updated, renamed, missing, skipped, and failed for plain ingest.ingested_production_roots plus the production document create/update/unchanged/retired and linked-part counters.created, updated, unchanged, retired, family reconstruction, linked images, and missing linked parts for ingest-production.npx claudepluginhub sdemyanov/retrieverIngests processed litigation and eDiscovery productions (DAT/OPT/TEXT/IMAGES sets, Bates-numbered volumes) via the Retriever tool template.
Seeds memory store from existing material (repos, PDFs, transcripts) via local or HTTP upload with preview, dry-run, and real ingestion modes.
Ingests PDF datasheets or reference manuals into the embedded docs search index via ingest_docs tool. Reports chunks ingested and tables found.