atlas-search
Skill0 starsChanged 10 days agoMITView on GitHub
Answer strictly from a live web search, never from internal knowledge or memory. Use when the user says "atlas search", "/atlas-search", "search the web", "actually look it up" — in any language — or asks for current facts where being out of date would matter. With "give me sources" in the request it returns where to read instead of the answer; "is there a free API / a skill / an MCP server for this" searches the public catalogues.
In this repository10 other files it gives its agents.
- atlas-artifactSkill
- atlas-commitSkill
- atlas-fixSkill
- atlas-helpSkill
- atlas-memorySkill
- atlas-organizeSkill
- atlas-recapSkill
- atlas-reviewSkill
- atlas-silentSkill
- atlasSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Do it here instead only when the caller has already fetched the page and is asking about something now in front of both of you.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Delegate to the atlas-research subagent and report what it returns.
The rules live there and only there: search first, never from memory, the date checked on anything that changes, two sources that disagree both reported, a gap named rather than filled. Repeating them here would mean two copies to keep aligned, and the copy is the one that goes stale.
What the delegation buys is where the pages land. Searching in this conversation leaves every fetched page in it for the rest of the session; the subagent reads them in a context that is discarded and hands back only the answer with its sources.
Do it here instead only when the caller has already fetched the page and is asking about something now in front of both of you.
A catalogue question goes to the same subagent, in its catalogue mode. "Is there a free API
for X", "what could host this for nothing", "is there already a skill / a plugin / an MCP server
that does this": first run the catalogue command here, where no web page is in context —
node "${CLAUDE_PLUGIN_ROOT}/tools/catalog.mjs" <kind> "<two or three words>" (kind: skills,
mcp, plugins, extras or all; at most eight lines per source) — then delegate to
atlas-research with those lines and the absolute path of search/catalog.md beside this
file (${CLAUDE_PLUGIN_ROOT}/skills/atlas-search/search/catalog.md), the file that holds the
sources and the checks. The agent reads pages and runs nothing. It finds; the install command
comes back as text, for the person to run.
A question about what people say goes to the same subagent too. Opinions, experiences, complaints, whether others hit the same problem, or a social network or forum named in the request, in any language: first run node "${CLAUDE_PLUGIN_ROOT}/tools/social.mjs" "<two or three words>" here, then delegate to atlas-research with those lines and the absolute path of search/social.md beside this file (${CLAUDE_PLUGIN_ROOT}/skills/atlas-search/search/social.md). The answer always ends by saying what was read, what was seen only as a snippet and what was not looked at.
This command means the answer, unless the request asks for sources. "Give me sources", "where can I read about this", "find me some videos", in any language: then it means the places to read — videos, discussions, articles, grouped by kind, one line each — and no answer. Same subagent, its other shape; the words in the request pick it, so nothing is guessed and nothing is asked.
The active compression level governs the prose here. It never overrides the format above: the shape stays, the words inside it shorten. A command that adds work is not permission to write long — what it adds is content, the prose around it is not.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

