‹ Back to apps
agentleFS
agentleFS plugins for Claude Code and Codex: connect your agent to permissioned context over MCP.
A community app, made by its publisher and not by the Grok Bot team. We scan apps before listing them, but we can’t see what an app’s server does with your data once you connect. Before you add an app
What’s inside
1 MCP server · 17 skills- MCP serveragentlefsMCP server: agentlefs
- Skillauthorization-modelHow agentleFS (afs) decides what a person or agent may read. Use when reasoning about afs access, grants, roles, cascade, groups or an agent's delegated scope; when a search or listing came back thinner than expected and you need to tell missing content from gated content; when explaining why someone or some agent cannot see a document; or when about to state who can or cannot see something. For carrying out a share, an access request or a who-can-see check, the sharing skill.
- Skillcatch-upCatch the user up on their work in agentleFS (afs) — what is waiting on them, what changed near their work, and where they left off. Use when they say \"catch me up\", \"what's waiting on me\", \"what changed\", \"what was I working on\", \"where did I leave off\" or \"get me up to speed\", or when they refer to their own ongoing work in afs (\"the pricing thing\", \"my spec\") as though you should already know which document they mean. Not for general questions, coding tasks in the open repository, or a session that has not touched afs.
- SkillcollaboratingCoordinate people and agents through agentleFS (afs): shared-document discussions, responding to review feedback or existing comment threads, direct questions and replies, findings, requests, handoffs, commitments, change subscriptions, document claims and scoped helper identities. Use documents for reviewing the source and anchoring feedback; use this skill for routing discussions and follow-up. For multi-agent waits, doorbell + filing-cabinet handoffs, and typed coordination also load swarm-board. Not for private chat advice with no shared-work context.
- SkillconnectionHow this client signs in to the agentleFS (afs) MCP server, and how to diagnose that connection. Use when agentleFS tools are unavailable, return 401, 404, or auth errors; when the user asks how to set up, sign in to, or authenticate this client with agentleFS; when the connection succeeds but returns no folders or reaches the wrong Organization or email; when configuring a self-hosted endpoint; or when setting up headless/CI access. Also use before concluding that agentleFS is broken. Syncing GitHub or Google Drive content into afs is the sources skill, not this one.
- SkilldeletingHow to delete, restore and permanently erase things in agentleFS (afs) without destroying more than intended. Use whenever a deletion is on the table — removing a document, clearing out a directory, deleting a folder, cleaning up after a migration or a bad ingest; when the user wants something deleted by mistake brought back or undone; when they ask to permanently erase or purge data, for example for a GDPR or other erasure request; and before calling doc_delete for any reason. Also use to explain why a delete or erase was refused, or why each needs two calls.
- SkilldocumentsFind, read, write, edit and review shared documents in agentleFS (afs). Use for team runbooks/specs/notes, document or folder browsing, review feedback, clarification questions, disagreements, and reading or responding to document comments—even when the user does not say comment. Also save topic summaries or this conversation. For recording one shared decision/fact use shared-truths-and-lessons; for team ballots use mandate-ledger. Folder documents (including wiki raw/ sources) belong here; synced GitHub/Drive connections use sources. Not for local repository files or general knowledge.
- Skillentry-mode-playbooksSoft recipes for unnamed harness entry modes (lost / ui / bug / feature) in agentleFS (afs) — assemble steps, caps, write-back checklist, and done-when. Loaded by proactive-assembly after soft mode inference; skippable and overrideable. Not a console wizard or entry-mode picker. Not for named continue-project or catch-up.
- Skillmandate-ledgerCreate, read and update team decision ballots in agentleFS (afs) and Mandate. Use when proposing or comparing approaches for a team-affecting product, architecture, rollout, pricing or policy choice; asking the user to choose between substantive alternatives; or asked to make a decision, seek team input or update a ballot. Also use when the user says a decision has been made, to verify the recorded choice and continue authorized ready work through decision pickup. Use shared-truths-and-lessons for facts and decisions already made without team review, and sharing for access requests.
- SkillmemoryRecall what earlier sessions saved to agentleFS (afs), such as session summaries, notes, decisions and lessons, when it would help the task at hand, and save a summary or note only when the user asks. Use when the user says \"remember this\", \"what did we do last time\", \"what did I save about X\", asks to keep something for next session, or asks how to make saving routine. Nothing is captured automatically; for one shared decision or a lesson use shared-truths-and-lessons, and for saving this conversation follow the documents skill's conversation-summary workflow.
- Skillproactive-assemblyWhen a coding or continue-work ask names no agentleFS (afs) project or folder (\"implement this feature\", \"fix the auth bug\", \"UI sucks\", \"don't know what to change\"), before coding infer a soft entry mode (lost / ui / bug / feature) and return one ranked, cited, permission-bound context packet — candidate projects/folders, docs/decisions, open asks/waits, suggested write-back targets. Suggestions only; capped; same reach as search/read. Skip when the user already named a project (continue-project). Not for catch-up, general Q&A, or when a project pointer is already present.
- SkillprojectsRun a project in agentleFS (afs) with other people and their agents, or alone. Use when the user wants to start a project (\"start a project to get us EU AI Act compliant\"), add people to one, see what is open or waiting in their projects, take on or hand back a piece of a project's work, report what they did, end a project, or split project work across helper agents (subagents, cloud sessions, `codex exec`) that report back.
- Skillshared-truths-and-lessonsRecord, check and challenge what an organization holds true, and keep lessons, in agentleFS (afs). Use when the user wants one specific decision, fact or assumption held as the organization's shared truth, with its evidence and rationale (\"record that we decided Postgres 16 is the minimum\", \"why did we choose this approach?\", \"save why we made this decision\"), says a recorded fact or decision is wrong and wants to push back on it, wants a lesson, workflow or dead end written down so nobody repeats it, or asks whether anyone — in their organization or in the public registry — has already shared a skill, workflow or lesson for something, including pulling one in or proposing one of theirs for publication. Not for a summary document of a topic or of several decisions (the documents skill) or saving this conversation (documents).
- SkillsharingGive, get and check access in agentleFS (afs). Use when the user wants to share a folder or document with a colleague, asks for access they do not have (\"I need access to the legal folder\"), has access requests waiting on their approval, asks who can read something or whether a particular agent can see something, or wants to offer a folder to someone in another organization or accept one offered to them. For why access works the way it does, the authorization-model skill.
- SkillsourcesBring content from other tools into agentleFS (afs) and check that it keeps arriving. Use when the user wants a GitHub repository or a Google Drive folder synced into afs so agents can search it (\"connect our Google Drive Specs folder\"), asks whether a connected source is syncing or why mirrored content is missing or stale, or asks whether afs can read a tool it has no connector for, such as Notion, Confluence or Jira. Also use when asked to link or bind a GitHub issue to an afs folder or document, resolve that link, or post its safe pointer on the issue. Not for documents already in an agentleFS folder, even when a task calls them \"sources\" (a wiki's raw/ sources, say), which are read with browse and read, the documents skill. Not for signing this client in to agentleFS itself, which is the connection skill.
- Skillswarm-boardSoft recipes so multi-agent or swarm work coordinates on agentleFS (afs) waits, typed messages, and shared docs — the board — instead of ad-hoc chat, prose task docs, or notebook locks. Load after connect when the user wants swarm handoffs, a Product→Admin doorbell, a shared filing cabinet across agent seats, spawning helpers, asking other agents, or waiting for a reply or a doc change. Especially under cron or intermittent heads, where waits were historically unused. Soft recipe, not a console wizard. Complements collaborating (primitives) and projects (folder briefs). Waits do not auto-start harnesses.
- Skilltake-a-shiftUse when the person asks 'any work?', 'what's waiting?', 'anything for my agents?', 'take a shift', or to pick up, continue as, work for or check on one of their agentleFS (afs) agents. Finds which of their agents have work waiting, clocks this session in on one whose seat is free, works its messages, and clocks out with a shift note so the next session in any tool (Claude Code, Cursor, Codex, Gemini CLI, Grok Build) takes the next shift. Also use when they say they are switching tools, ending the session or running low on tokens, when they ask to put a number of agents on a folder, or to stop, check on or resume that crew, and when they want a colleague to take shifts on a folder's agents.
- Skillwriteback-loopBefore stopping an unnamed or lightly named agentleFS (afs) harness session that assembled context or produced durable findings, run a guided write-back checklist — commit docs/comments/messages/truths to the suggested targets (or an agreed folder) using existing MCP write tools. Soft recipe, not a wizard; human can skip. Product form of VF-auto-writeback. Not for sessions that already wrote back, pure chat with nothing durable, or when the human said not to record.