How do I install the Query Companies skill?

Run npx skillmds@latest add brianbolze/query-companies 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 Query Companies 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 Query Companies?

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 Query Companies free to use?

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

Who published Query Companies?

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

Query Companies by brianbolze — Agent Skill

Name: Query Companies
Author: brianbolze

Query Companies

Use for named-company comparison or lookup prompts that can be answered from the local web-research company store, especially pricing, offerings, ownership, catalog breadth, cohort cuts, price-visibility, and broad single-company briefs: "tell me about X", "compare X and Y", "what do these competitors charge / offer", "which brands sell <thing>", "who owns X", "is X in the web-research store", "what has the store captured on X". Answers are captured-state snapshots from cited primary-source dossiers instead of WebSearch. Do not browse, WebSearch, curl, or open live company pages just to verify; resolve with store.py find, answer from store files, and cite local paths plus governing capture clocks. Read-only: never scrapes, never spends. If current/latest or missing/stale capture is required, suggest /research-company refresh; for incomplete rosters suggest /deepen-offerings; for external signals use tools/. Not for general web search, news/funding events, financials, judgments, or non-company topics.

brianbolze 3835562 2 files · 3.9 KB Updated Sep 17, 2026

File contents

/query-companies — answer from the company store, not the live web

Router + guardrails over QUERYING.md. Read-only; never scrapes, never spends. Recipes live in QUERYING.md; this file only fixes order of operations and points at the trust rules.

Steps

Resolve the store: $WEB_RESEARCH_HOME first; if unset (harnesses that skip the shell profile), fall back to the canonical single-user path "/Users/brianbolze/Library/Mobile Documents/com~apple~CloudDocs/Web Research". Quote it — the path has spaces.
Resolve names -> slugs, always: python "$WEB_RESEARCH_HOME/scripts/store.py" find <each>. Report per-company status before answering: clocks per layer / STUB / not in store. If find prints likely candidate, inspect that slug before calling it a miss. If it prints ambiguous candidates, do not pick silently: use nearby context only if it clearly disambiguates; otherwise ask the user which company they meant, listing the candidate slugs/names.
Route via QUERYING.md's one-rule table: for a human-facing ask about one profiled company, run render.py and paste the link line it prints into your reply (rendering without linking buries the artifact); rg for locate; PyYAML parse for structure; store.db only for many-pivot asks, and rebuild it first.
Group answers wear the §0 stamp: if the answer covers more than one company, open with QUERYING.md §0's one-line stamp (Group · Set · Leaves out · Claim). The set-type caps the claim; a question that needs more than the type allows becomes "open question," not a clean number.
Answer captured state from store files only. No WebSearch/browser/curl. Cite path
- governing clock; prices use the offerings clock when they come from offerings.md.
Gaps are hand-offs, never silent live fallback: cold -> suggest /research-company; thin roster -> /deepen-offerings; external signal -> the specific tools/*.py.

Trust Rules

Negatives: apply QUERYING.md's before-trusting-a-negative check first.
Counts: enumeration-gated (QUERYING.md Recipe 4); floors are floors.
Prices: verbatim + intra-cohort + hand-normalized; never a sorted $ / mo table.
Cohort cuts: unclear/empty = "couldn't tell," not "no."
Groups: every multi-company answer opens with the §0 stamp; the set-type caps the claim (store coverage ≠ market; empty field = "at least N," not "the rest don't"; outside-list miss = "not in our store," not "not in the market").

Scope

State only. Events, judgments, and financials are out of store scope; say so.

---name: query-companiesdescription: Use for named-company comparison or lookup prompts that can be answered from the local web-research company store, especially pricing, offerings, ownership, catalog breadth, cohort cuts, price-visibility, and broad single-company briefs: "tell me about X", "compare X and Y", "what do these competitors charge / offer", "which brands sell <thing>", "who owns X", "is X in the web-research store", "what has the store captured on X". Answers are captured-state snapshots from cited primary-source dossiers instead of WebSearch. Do not browse, WebSearch, curl, or open live company pages just to verify; resolve with store.py find, answer from store files, and cite local paths plus governing capture clocks. Read-only: never scrapes, never spends. If current/latest or missing/stale capture is required, suggest /research-company refresh; for incomplete rosters suggest /deepen-offerings; for external signals use tools/. Not for general web search, news/funding events, financials, judgments, or non-company topics.---# /query-companies — answer from the company store, not the live webRouter + guardrails over `QUERYING.md`. Read-only; never scrapes, never spends.Recipes live in `QUERYING.md`; this file only fixes order of operations andpoints at the trust rules.## Steps1. Resolve the store: `$WEB_RESEARCH_HOME` first; if unset (harnesses that skip the shell profile), fall back to the canonical single-user path `"/Users/brianbolze/Library/Mobile Documents/com~apple~CloudDocs/Web Research"`. Quote it — the path has spaces.2. Resolve names -> slugs, always: `python "$WEB_RESEARCH_HOME/scripts/store.py" find <each>`. Report per-company status before answering: clocks per layer / `STUB` / not in store. If `find` prints `likely candidate`, inspect that slug before calling it a miss. If it prints `ambiguous candidates`, do not pick silently: use nearby context only if it clearly disambiguates; otherwise ask the user which company they meant, listing the candidate slugs/names.3. Route via `QUERYING.md`'s one-rule table: for a human-facing ask about one profiled company, run `render.py` and **paste the link line it prints into your reply** (rendering without linking buries the artifact); `rg` for locate; PyYAML parse for structure; `store.db` only for many-pivot asks, and rebuild it first.4. Group answers wear the §0 stamp: if the answer covers more than one company, open with `QUERYING.md` §0's one-line stamp (`Group · Set · Leaves out · Claim`). The set-type caps the claim; a question that needs more than the type allows becomes "open question," not a clean number.5. Answer captured state from store files only. No WebSearch/browser/curl. Cite path + governing clock; prices use the offerings clock when they come from `offerings.md`.6. Gaps are hand-offs, never silent live fallback: cold -> suggest `/research-company`; thin roster -> `/deepen-offerings`; external signal -> the specific `tools/*.py`.## Trust Rules- Negatives: apply `QUERYING.md`'s before-trusting-a-negative check first.- Counts: enumeration-gated (`QUERYING.md` Recipe 4); floors are floors.- Prices: verbatim + intra-cohort + hand-normalized; never a sorted `$ / mo` table.- Cohort cuts: unclear/empty = "couldn't tell," not "no."- Groups: every multi-company answer opens with the §0 stamp; the set-type caps the claim (store coverage ≠ market; empty field = "at least N," not "the rest don't"; outside-list miss = "not in our store," not "not in the market").## ScopeState only. Events, judgments, and financials are out of store scope; say so.

Frequently asked questions