How do I install the Drive File Search skill?

Run npx skillmds@latest add vamseeachanta/drive-file-search in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Drive File Search safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Drive File Search?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Drive File Search free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Drive File Search?

vamseeachanta (@vamseeachanta) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Drive File Search by vamseeachanta — Agent Skill

Name: Drive File Search
Author: vamseeachanta

Drive File Search

Surface SPECIFIC FILES on the shared drives (/mnt/ace, /mnt/dde) relevant to the current task — the FILE-level complement to ecosystem-data-sources (domain-catalog level). Extracts query terms from work context, runs the drive-index CLI once, presents ranked canonical paths with freshness caveats. Use for similar-past-work, prior-project, or files/examples/precedent asks; when domain work needs past deliverables (mooring calc → xlsx/py; CAD → dwg/step; standards → PDF inventory).

vamseeachanta Updated Sep 17, 2026

File contents

Drive File Search — context-aware related-file surfacing

FILE-level asks only; domain-catalog asks belong to ecosystem-data-sources.

Procedure

Extract context (full detail + worked walkthrough: references/context-extraction.md). Signals: repo; issue title/labels; domain via keywords read live from .claude/hooks/ecosystem-domain-map.json (generated — never hand-copy its keywords here); artifact-type hints (calcs → xlsx/py/csv; CAD → dwg/ipt/step; standards → pdf; simulation → dat/sim/yml/owr).
Build the query — 2–5 deduped terms (domain keyword + 1–3 title words). Pass --domain <d> only if <d> appears in a domains: list in config/drive-index-registry.yml; else fold the domain word into the terms (unknown --domain = empty selection, exit 0).
Run the ONE command (read-only):

uv run python scripts/data/drive-index-search/search.py "<terms>" [--domain <d>] --json --limit 20 --caller skill

(python3 if uv is broken.)
Branch on exit code:
- 0 — parse the JSON envelope (covers partial and empty results — always check coverage_gaps).
- 2 — registry error OR zero reachable indexes: name the down indexes/drives, point to scripts/setup/canonical-drive-links.sh (PR #3341 if absent), offer ecosystem-data-sources as fallback.
- CLI missing — cite issue #3335, fall back to ecosystem-data-sources; never run ad-hoc drive crawls (bounded reads only).
Present top 10. Per result: canonical_path, source_index, score, rank_basis (meta.* optional; raw_path never shown). Show the canonical /mnt/... path; why relevant: matched terms, rank_basis, extension-vs-hint (reorder within equal scores only); source index + freshness caveat from the registry's freshness/built_at ("freshness unknown" if absent). Quote each coverage_gaps reason VERBATIM (opaque free text — never paraphrase); name the excluded drive. Offer three next actions: open one listed file (bounded read); record chosen paths under "Documents consulted" in the plan's Resource Intel section; refine terms, re-run once.

Guardrails (never violate)

Read-only: never write to /mnt/ace or /mnt/dde; the CLI opens indexes read-only.
De-identification (public repo): drive paths embed client names/project codes. Before quoting a surfaced path into an issue, plan, PR, or commit, scan for client tokens; if found, describe as metadata only ("a past deliverable on /mnt/ace") or redact. In-session display is fine; persisting into public artifacts is the gated act. De-id stays on lane:claude.
Unreachable drive is normal: present partial results plus the coverage_gaps caveat; never mount, never sudo.
Usage playbook (#3340): integration points, freshness authority, metrics, and de-id checklist: docs/guides/drive-file-search-playbook.md.

Related: ecosystem-data-sources (DATA/domain-catalog level).

---name: drive-file-searchdescription: Surface SPECIFIC FILES on the shared drives (/mnt/ace, /mnt/dde) relevant to the current task — the FILE-level complement to ecosystem-data-sources (domain-catalog level). Extracts query terms from work context, runs the drive-index CLI once, presents ranked canonical paths with freshness caveats. Use for similar-past-work, prior-project, or files/examples/precedent asks; when domain work needs past deliverables (mooring calc → xlsx/py; CAD → dwg/step; standards → PDF inventory).---# Drive File Search — context-aware related-file surfacingFILE-level asks only; domain-catalog asks belong to `ecosystem-data-sources`.## Procedure1. **Extract context** (full detail + worked walkthrough: `references/context-extraction.md`). Signals: repo; issue title/labels; domain via keywords read live from `.claude/hooks/ecosystem-domain-map.json` (generated — never hand-copy its keywords here); artifact-type hints (calcs → xlsx/py/csv; CAD → dwg/ipt/step; standards → pdf; simulation → dat/sim/yml/owr).2. **Build the query** — 2–5 deduped terms (domain keyword + 1–3 title words). Pass `--domain <d>` only if `<d>` appears in a `domains:` list in `config/drive-index-registry.yml`; else fold the domain word into the terms (unknown `--domain` = empty selection, exit 0).3. **Run the ONE command** (read-only): `uv run python scripts/data/drive-index-search/search.py "<terms>" [--domain <d>] --json --limit 20 --caller skill` (`python3` if uv is broken.)4. **Branch on exit code**: - `0` — parse the JSON envelope (covers partial and empty results — always check `coverage_gaps`). - `2` — registry error OR zero reachable indexes: name the down indexes/drives, point to `scripts/setup/canonical-drive-links.sh` (PR #3341 if absent), offer `ecosystem-data-sources` as fallback. - CLI missing — cite issue #3335, fall back to `ecosystem-data-sources`; never run ad-hoc drive crawls (bounded reads only).5. **Present top 10**. Per result: `canonical_path`, `source_index`, `score`, `rank_basis` (`meta.*` optional; `raw_path` never shown). Show the canonical `/mnt/...` path; why relevant: matched terms, `rank_basis`, extension-vs-hint (reorder within equal scores only); source index + freshness caveat from the registry's `freshness/built_at` ("freshness unknown" if absent). Quote each `coverage_gaps` `reason` VERBATIM (opaque free text — never paraphrase); name the excluded drive. Offer three next actions: open one listed file (bounded read); record chosen paths under "Documents consulted" in the plan's Resource Intel section; refine terms, re-run once.## Guardrails (never violate)- **Read-only**: never write to `/mnt/ace` or `/mnt/dde`; the CLI opens indexes read-only.- **De-identification** (public repo): drive paths embed client names/project codes. Before quoting a surfaced path into an issue, plan, PR, or commit, scan for client tokens; if found, describe as metadata only ("a past <domain> deliverable on /mnt/ace") or redact. In-session display is fine; persisting into public artifacts is the gated act. De-id stays on lane:claude.- **Unreachable drive is normal**: present partial results plus the coverage_gaps caveat; never mount, never sudo.- **Usage playbook** (#3340): integration points, freshness authority, metrics, and de-id checklist: `docs/guides/drive-file-search-playbook.md`.**Related**: `ecosystem-data-sources` (DATA/domain-catalog level).

Frequently asked questions