Atlas / MCP servers / zibbydev / @zibby/skills

@zibby/skillsBLOCK

Name: @zibby/skills
Author: zibbydev

mcp/zibbydev/skills

MCP server by zibbydev · agent skills · ai agents · automation · Audited 2026-10-12

Pluggable MCP tool modules that give agent workflows their capabilities — GitHub, GitLab, Sentry, Slack, Jira, Notion, memory, datasets, and more.

Verdict

BLOCK

Grade

Trust score

42 /100

Exposed tools

200 159r · 81w · 11d

Transport

stdio · streamable-http

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

[](https://www.npmjs.com/package/@zibby/skills) [](https://www.npmjs.com/package/@zibby/skills) [](./LICENSE)

📖 Full docs: docs.zibby.app · Get Started · Concepts · Designing agents · CLI Reference · Cloud

The skill layer for [@zibby/agent-workflow](https://github.com/ZibbyDev/agent-workflow). Built-in skill definitions that give a workflow node the tools it needs — function tools, MCP servers, browser, issue trackers, memory. Vendor-neutral, JavaScript-first.

@zibby/skills is the batteries-included companion to `@zibby/agent-workflow` (npm) — "Graph-based AI agent workflow orchestration." The workflow engine ships zero skills on purpose; this package is where the built-in ones live.

A skill is the contract between a workflow node and a tool. It tells the engine what the tool does, how to start it, and what it needs. The engine never hardcodes any skill by name — it reads the skill definition and wires things up generically for both Claude and Cursor agents.

@zibby/agent-workflow node            @zibby/skills
──────────────────────────            ─────────────
skills: ['add']            ──►        getSkill('add')

Read from source at commit 0ee40fe4a03bOBSERVED · 2026-10-12

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add pingcode-mcp --env DISCORD_BOT_TOKEN=${DISCORD_BOT_TOKEN} --env GITHUB_TOKEN=${GITHUB_TOKEN} --env GITLAB_OAUTH_TOKEN=${GITLAB_OAUTH_TOKEN} --env GITLAB_TOKEN=${GITLAB_TOKEN} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "pingcode-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "DISCORD_BOT_TOKEN": "${DISCORD_BOT_TOKEN}",
        "GITHUB_TOKEN": "${GITHUB_TOKEN}",
        "GITLAB_OAUTH_TOKEN": "${GITLAB_OAUTH_TOKEN}",
        "GITLAB_TOKEN": "${GITLAB_TOKEN}"
      }
    }
  }
}

Exposed tools (200)

159 read · 81 write · 11 destructive. Blast radius: 11 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`DM`	read
`Eng`	read	engineering
`On-Call`	read	Weekly rotation
`add_node`	write	Add a new node to an existing workflow. Generates the node file and updates the barrel export.
`artifact_get`	read	Fetch ONE artifact you published, by id → { metadata, content }. Use to reuse / edit / re-publish a page. To LIST what you have published, use your kv-memory tool kv_recall_prefix with keyPrefix
`artifact_publish`	write	Publish a NEW self-contained, shareable page (report/plan/table/dashboard/diagram/write-up) and get back a shareable URL.
`artifact_share`	read	Make an artifact viewable WITHOUT a Zibby login, by password. Returns { url, password } — a link anyone can open plus the password that opens it.
`artifact_unshare`	destructive	REMOVE an artifact\
`artifact_update`	write	Revise an EXISTING artifact by id — the shareable URL stays the same, the content is replaced (new version). Pass the fields to change (title and/or markdown\|html\|text).
`auth_status`	read
`build_workflow`	read	Generate real workflow code on disk from a design spec. Uses the configured AI agent for high-quality code generation.
`chart_render`	read	Render a chart LOCALLY (no external service) from a standard Apache ECharts option.
`close_workspace`	destructive	Remove an unused clean workspace and release its storage reservation. Refuses to discard changes, local commits or extra files. Does not delete the original host directory.
`code_stats_glob_hit`	read	How many of paths match ANY of globs (supports *, , ?). Returns { total, hits, ratio, matched }. Use for reproducible core-path coverage against the caller\
`code_stats_percentile`	read	The 0-100 percentile rank of value within values (cumulative
`create_ticket`	write
`dataset_query`	write	Run a SQL-style query over a bound store to build reports: select/aggregate (count\|sum\|avg\|min\|max), filter, group, order, limit, and bound by month. Returns { columns, rows }. Use this to compute summaries/aggregations from records you appended earlier.
`deploy_workflow`	write	Deploy a workflow to Zibby Cloud. Returns the trigger URL.
`design_workflow`	read	Design a workflow spec (nodes, edges, descriptions) for the user to review before building. Call this after understanding requirements.
`discord_list_channels`	read	List the text channels of the connected Discord server (id + name). Pass guildId only when the bot is in multiple servers.
`discord_send_message`	write	Post a message to a Discord channel as the connected bot. Text over 2000 chars is automatically split into sequential messages.
`echo`	read	Echo one message
`ensure_store`	write	Create (or reuse) a store ON DEMAND for THIS agent — use when you need a store that was NOT declared/bound at deploy. Idempotent by name and private to this agent: calling again with the same name returns the SAME store (safe to call at the start of every run). Returns { storeId }. For type
`explore_framework_docs`	read	Read Zibby framework documentation on demand. Call this before building complex workflows or when you need details on advanced patterns (middleware, conditional routing, skills, deployment, CLI commands).
`figma_get_comments`	read	Read the comments on a Figma file
`figma_get_file`	read	Get a Figma file\
`figma_get_file_versions`	read	List a Figma file\
`figma_get_image_fills`	read	Get a map of imageRef → download URL for every image fill used in a Figma file.
`figma_get_me`	read	Get the authenticated Figma user profile (handle, email, id) — confirms the token works.
`figma_get_nodes`	read	Get specific nodes from a Figma file by their node ids. Use after figma_get_file to inspect a particular frame/component without re-fetching the whole file. Node ids look like
`figma_get_project_files`	read	List the files in a Figma project (from figma_get_team_projects). Returns file key, name, thumbnail, last_modified.
`figma_get_team_projects`	read	List the projects in a Figma team. Get team_id from a Figma team URL (.../files/team/<TEAM_ID>/...).
`figma_parse_url`	read	Local helper (no API call): extract the file key and node id (colon form) from a figma.com file/design URL. Returns { file_key, node_id }.
`figma_post_comment`	write	Post a comment on a Figma file (WRITE — needs a PAT with comment-write scope). Optionally reply to a comment (comment_id), pin it to a node (node_id, with optional x/y offset), or pin to a canvas point (x/y).
`figma_render_png`	read	Render one or more Figma nodes to an image and return the URLs (a map of nodeId → URL). Despite the name it renders PNG/JPG/SVG/PDF (default png). Use to show or download a visual of a frame/component.
`file_delete`	destructive	For FILE-type stores: delete ONE stored file by its relative path. Returns { ok, path, deleted }. Errors if the path does not exist.
`file_get`	read	For FILE-type stores: read ONE stored file back by its relative path. Text files come back as
`file_list`	read	For FILE-type stores: list the stored files — each entry has { path, size, lastModified }. Optionally filter by a path
`file_put`	destructive	For FILE-type stores: write (or overwrite) ONE file at a relative path — raw JSON dumps, CSVs, images, snapshots, any blob you want to keep across your stateless runs. Text goes in
`gbrain_delete`	destructive	Remove documents from the knowledge base by their stable sourceId(s).
`gbrain_ingest`	read	Upsert documents into the knowledge base. Each doc is { sourceId, markdown, deleted? }; sourceId is a STABLE id (e.g.
`gbrain_query`	read	Semantic-search the knowledge base and return the most relevant document chunks (each with its sourceId and relevance score).
`gdocs_append`	read	Append markdown/text content to the END of an existing Google Doc. Only works on docs this app created or the user explicitly picked (drive.file scope). Accepts a documentId or a full docs.google.com URL. Returns { ok, documentId, url }.
`gdocs_create_doc`	write	Create a new Google Doc with a title and optional content (markdown: #/##/### headings, - bullets, 1. numbered lists, bold, [links](url),
`gdocs_insert_image`	write	Append a LOCAL image file (png/jpg, max 5MB) to the END of an existing Google Doc. The image is uploaded to the user\
`gdocs_list_created`	read	List the Google Docs visible to this integration, newest first (max 25). NOTE: under the drive.file scope this lists ONLY docs the app created or the user explicitly picked — it is NOT a full Drive search. Returns { ok, count, files:[{ documentId, title, modifiedTime, url }] }.
`git_checkout`	read	Clone a git repository locally (or pull latest if already cloned). Auto-authenticates with GitHub/GitLab tokens if available.
`git_explore`	read	Quick structural overview of a cloned repo: key files, package.json info, directory tree (top 2 levels), detected framework/language
`git_list_repos`	read	List locally cloned repositories
`github_add_issue_comment`	write	Add a comment to a GitHub issue. Also the way to record a PR link on an issue (post a markdown link).
`github_clone`	read	Clone a GitHub repository to the local filesystem. Use when user says
`github_close_issue`	write	Close a GitHub issue. Optionally set the close reason (completed or not_planned).
`github_create_issue`	write	Create a GitHub issue
`github_create_review`	write	Post a review on a pull request: a summary body plus optional inline comments anchored to file/line, with an event (COMMENT, APPROVE, or REQUEST_CHANGES). Use this to deliver a code review back to the PR.
`github_get_commit`	write	Get details of a specific commit — message, stats, file diffs
`github_get_file`	read	Read a file (or list a directory) from a GitHub repo, without cloning. Works on any branch/ref.
`github_get_issue`	read	Get a single GitHub issue with full detail (title, body, state, labels, assignee, url)
`github_get_issue_comments`	read	Get the comment thread on a GitHub issue (chronological)
`github_get_pr`	read	Get details of a pull request — title, description, branch, stats
`github_get_pr_diff`	read	Get the unified diff of a pull request — the actual code changes
`github_get_review_thread`	read	Read a PR review-comment THREAD given any comment id in it: the root review comment + all its replies, plus the anchored diff context (file, line, the original diff hunk). Use this to understand a human\
`github_get_user`	read	Get the authenticated GitHub user profile and their organizations
`github_label_issue`	destructive	Add, set (replace all), or remove labels on a GitHub issue. Labels back state-like transitions on GitHub.
`github_list_commits`	read	List commits on a branch — optionally bounded by since/until ISO dates (date-window listing) and paginated (keep calling with page+1 while hasMore is true). Optionally filtered by file path.
`github_list_issues`	read	List issues in a repo (excludes pull requests). Filter by state, labels, and an updated-since cursor for polling.
`github_list_orgs`	read	List GitHub organizations the authenticated user belongs to
`github_list_pr_comments`	read	Get all review and issue comments on a PR
`github_list_pr_files`	read	List files changed in a PR with per-file patches
`github_list_repos`	read	List the repositories this token/installation can access (omit owner) — or a specific user/org\
`github_reopen_issue`	read	Reopen a closed GitHub issue
`github_reply_issue_comment`	write	Post a reply on a PR\
`github_reply_review_thread`	read	Reply IN-THREAD to an existing PR review-comment thread (a conversational reply nested under the thread the human commented on — NOT a fresh full review). Pass any comment id in the thread.
`github_search_code`	read	Search code across GitHub repositories by keyword
`github_search_issues`	read	Search GitHub issues and pull requests
`github_search_repos`	read	Search accessible repositories by name or description. Use this when the user asks to find a specific repo.
`gitlab_add_issue_comment`	write	Add a comment to a GitLab issue. Also the way to record an MR link on a ticket (post a markdown link).
`gitlab_create_mr_review`	write	Post a full review on a GitLab merge request in one call: a summary note plus optional inline comments anchored to file/line in the diff. Convenience wrapper over gitlab_post_mr_note + gitlab_post_mr_discussion.
`gitlab_get_commit`	write	Get ONE commit\
`gitlab_get_discussion`	write	Read a single GitLab merge-request DISCUSSION (thread) by its discussion id: all notes in order plus the diff position (file + line) it is anchored to. Use this to understand a human\
`gitlab_get_issue`	read	Get a single GitLab issue with full detail (title, description, state, labels, assignees, web url).
`gitlab_get_mr_changes`	write	Get the changed files of a GitLab merge request with per-file diffs — the actual code changes to review. Also returns diff_refs for inline comments.
`gitlab_list_commits`	read	List a project\
`gitlab_list_issues`	read	List a GitLab project\
`gitlab_list_mr_notes`	write	List the discussion notes on a GitLab merge request (chronological).
`gitlab_list_mrs`	read	List a GitLab project\
`gitlab_list_projects`	read	List the GitLab projects this token can access (the projects you are a member of), optionally filtered by a search query. Use this to discover a RELATED project worth cloning when a change\
`gitlab_post_mr_note`	write	Post a general (non-inline) comment on a GitLab merge request. Use for a review summary or a top-level remark.
`gitlab_reply_discussion`	write	Reply IN-THREAD to an existing GitLab merge-request discussion (a conversational reply appended to the SAME thread — NOT a fresh review). Use after gitlab_get_discussion to answer a human\
`graph_get`	read	Fetch one node (by
`graph_put`	write	Create or update a node (an entity: a file, ticket, run, person, repo, conclusion) in the memory graph. Upsert by id (
`graph_subgraph`	read	Export the induced subgraph around a query: every node the same recall would reach, plus ALL live edges among those nodes (not just the cheapest paths). The input for visualisation and for hand-offs that need the full local structure. Same filters as a recall query; the seeds are always included.
`graph_trace`	read	A node
`hubspot_associate`	read	Associate two CRM records with the default association type (e.g. a contact to a company). Returns { ok:true }.
`hubspot_create_note`	write	Log a note (hs_note_body) and optionally associate it to a contact/company/deal/ticket. Returns { id, associatedTo }.
`hubspot_create_object`	write	Create a CRM record. Pass
`hubspot_get_object`	read	Fetch one CRM record by id. Returns { id, properties }.
`hubspot_list_objects`	read	Page through records of a CRM object type. Returns { results, paging } — pass paging.next.after back as
`hubspot_list_properties`	read	List the defined properties (name, label, type, fieldType) for a CRM object type. Call this FIRST so you use valid property names when creating/updating/searching.
`hubspot_search`	read	Search a CRM object type. Pass a full-text
`hubspot_update_object`	write	Update (PATCH) a CRM record by id. Pass
`hubspot_whoami`	read	Confirm the HubSpot token works: returns the portal (hub) id, account type, time zone and UI domain.
`install_skill`	write	Install a skill into the current chat session so its tools become available
`jira_add_comment`	write	Add a comment to a Jira issue
`jira_create_issue`	write	Create a new Jira issue. For
`jira_edit_issue`	write	Update fields on a Jira issue (summary, description, story points, labels, priority). A string
`jira_get_comments`	read	Get comments on a Jira issue (newest first) — use this to find testing steps, notes, etc. When the user asks about testing steps, test cases, or wants to run tests for a ticket, call this: testing steps are typically written in the ticket\
`jira_get_issue`	read	Get details of a specific Jira issue
`jira_get_sprint_issues`	read	Get all issues in a sprint, optionally filtered by status column name. Returns status breakdown.
`jira_list_issue_types`	read	List issue types allowed for issue creation in the given project.
`jira_list_projects`	read	List all Jira projects accessible to the user. Call this when the user asks about projects or boards.
`jira_list_sprints`	read	List sprints for a Jira project (returns sprint names, IDs, states, dates). When the user asks about sprints: jira_list_sprints → jira_get_sprint_issues.
`jira_list_statuses`	read	List Jira statuses. Use projectKey to get statuses applicable in that project workflow.
`jira_move_issue_to_sprint`	write	Move an issue to a sprint by id/name/target and verify membership. When user asks to move ticket into a sprint, do NOT use status transition: use this tool (issueKey, projectKey?, sprintId\|sprintName\|target) and report verified result.
`jira_move_to_active_sprint`	write	Backward-compatible alias: move issue to sprint target and verify membership.
`jira_search`	read	Search Jira issues using JQL.
`jira_transition_issue`	write	Move a Jira issue to a different status. Always pass toStatus when user gave a target; only pass issueKey alone when you explicitly need to list transitions.
`kv_recall`	read	Recall the value you stored under a plain key (exact match). Your per-agent namespace is added automatically — pass a plain key like
`kv_recall_prefix`	read	List your entries whose plain key STARTS WITH a prefix (e.g.
`kv_store`	destructive	Store (overwrite) a value under a plain key so a later run of yours can recall it. Your per-agent namespace is added automatically.
`lark_get_chat_history`	read
`lark_list_chats`	read
`lark_lookup_user_by_email`	read
`lark_reply`	read
`lark_search_users`	read
`lark_send_message`	write
`larkattendance_list_groups`	write	DISCOVERY (start here). List this tenant\
`larkattendance_list_stats_fields`	read	DISCOVERY — call this BEFORE larkattendance_query_stats. Lists every statistic column this tenant has for the period, grouped (e.g. 基本信息 / 出勤统计 / 异常统计 / 每日统计 / 自定义字段). The 自定义字段 group is tenant-defined and exists nowhere else, so this is the only way to learn its codes. Match the user\
`larkattendance_resolve_users`	read	Map work EMAIL addresses to the attendance user ids (tenant user_id = the attendance API\
`larkbitable_read_records`	read	Read the rows of one table in a Lark/Feishu Base (多维表格 / bitable) as flat text. Accepts a Base URL (/base/... or a /wiki/... link fronting a Base) or app token; the table and view are taken from the URL
`larkdoc_add_comment`	write	Post a NEW top-level comment on a Lark/Feishu document. Accepts a documentId or full doc URL plus text. Returns { ok, documentId, commentId }.
`larkdoc_append`	read	Append markdown/text content to the END of an existing Lark/Feishu document (docx). Accepts a documentId or a full Lark doc URL. Returns { ok, documentId, url }.
`larkdoc_get`	read	Read a Lark/Feishu document (docx) as plain text, for use as reference context. Accepts a raw doc id OR a full Lark doc URL (a /docx/ or /wiki/ link — wiki links are resolved to their backing docx). Returns { ok, documentId, title, url, text }. Text is truncated to ~20k chars.
`larkdoc_insert_image`	write	Append a LOCAL image file (png/jpg, max 10MB) to the END of an existing Lark/Feishu document (docx). Accepts a documentId or full doc URL plus a local file path. Optional width/height in pixels. Returns { ok, documentId, blockId, fileToken, url }.
`larkdoc_list_comments`	read	List the comment threads on a Lark/Feishu document. Accepts a documentId or full doc URL. Returns { ok, comments:[{ commentId, replies:[{ replyId, author, text }] }] }. Use to read the thread you are replying to.
`larkdoc_reply_comment`	read	Reply INSIDE an existing comment thread on a Lark/Feishu document. Pass { documentId, commentId, text }. Use to answer a user who @mentioned Zibby in a doc comment (reply in the same commentId). Returns { ok, documentId, commentId, replyId }.
`larkwiki_list_spaces`	read	List the Lark/Feishu wiki spaces visible to the connected app. Returns { ok, count, spaces:[{ spaceId, name }] }. Use to discover the wikiSpaceId for larkdoc_create.
`linear_add_comment`	write	Add a comment to a Linear issue (markdown supported).
`linear_get_comments`	read	Get comments on a Linear issue (newest first).
`linear_get_issue`	read	Get a single Linear issue by identifier (e.g. ENG-123) or internal uuid — title, description, state, labels, assignee, url.
`linear_link_attachment`	read	Attach a URL (e.g. a GitHub PR) to a Linear issue via native attachments. Use this for PR links; fall back to linear_add_comment if it fails.
`linear_list_issues`	read	List/poll Linear issues filtered by team, state, label, assignee, and an updatedAfter cursor. Returns newest-updated first.
`linear_list_labels`	read	List issue labels, optionally scoped to a team.
`linear_list_states`	read	List a team
`linear_list_teams`	read	List Linear teams (id, key, name). Needed to scope workflow states and issue queries.
`linear_update_state`	write	Move a Linear issue to a different workflow state. Pass a state NAME (toStatus/stateName) and the tool resolves it to the issue
`linkedin_list_organizations`	read	List the LinkedIn Organizations (company Pages) the authenticated member ADMINISTERS. Returns [{ id, urn, name, vanityName }]. Call this first to choose the author org for a draft post.
`linkedin_publish_post`	write	PUBLISH a post to the authenticated member\
`list_available_skills`	read	List all skills that can be installed, with their env-var readiness status
`list_directory`	read	List files and directories in a path. Returns names with type indicators (/ for dirs).
`list_messages`	read	List YOUR agent\
`list_products`	read
`list_running_agents`	write	List the runs active right now that this run may reach (the agent\
`list_specs`	read	List available test spec files in the project
`list_ticket_types`	read
`list_tickets`	read
`list_workflows`	read	List all local workflows in .zibby/workflows/.
`list_workspaces`	read	List this conversation’s saved workspaces, status, revisions, directories and account storage reservations. Does not start another agent.
`memory_brief`	read	Get a compact briefing: recent session summaries + top relevant facts. Call at the start of a conversation.
`memory_end_session`	write	End the current session and save a summary for future recall. Call when a task is complete.
`memory_get_navigation`	read	Query known page-to-page transitions
`memory_get_page_model`	read	Query page structure — elements, roles, selectors
`memory_get_selectors`	read	Query known selectors for a page with stability metrics
`memory_get_test_history`	read	Query recent test runs with pass/fail results and timing
`memory_recall`	read	Search persistent memory by keyword, category, ticket, or tier. Returns matching facts and context.
`memory_save_insight`	write	Save a useful observation for future runs (selector tips, timing, workarounds)
`memory_store`	write	Save a fact, decision, or context to persistent memory. Survives across sessions.
`message_agent`	read	Leave a note for another agent — OR FOR YOURSELF, to be picked up in a later round.
`needs_you`	write	Put a card in front of a PERSON on the project\
`notion_add_comment`	write	Post a comment on Notion. To REPLY within an existing discussion, pass { discussionId, text }. To start a NEW top-level comment on a page, pass { pageId, text }. Returns { ok, id, discussionId }. Use this to answer a user who @mentioned Zibby in a Notion comment (reply in the same discussionId).
`notion_append_blocks`	read	Append markdown (or raw Notion blocks) to the END of an existing Notion page. Accepts a page id or full Notion URL. Chunks automatically at Notion\
`notion_get_page`	read	Fetch a Notion page and its content flattened to markdown, for use as read-only context. Accepts a raw page id OR a full Notion URL. Returns { ok, id, title, url, text }. Text is truncated to ~20k chars.
`notion_insert_image`	write	Append an image block to a Notion page. Pass imagePath for a LOCAL .png/.jpg file (max 20MB — uploaded via Notion
`notion_list_comments`	read	List the open comment discussions on a Notion page/block. Accepts a page/block id OR a full Notion URL. Returns { ok, comments:[{ id, discussionId, text, author }] }. Use to read the comment thread you are replying to.
`notion_query_database`	read	Query a Notion database and return a bounded list of rows (id, title, url, key props). Accepts a database id OR full Notion URL. Optional Notion filter object. Returns at most 25 rows. Use to find a specific page, then notion_get_page it.
`open_url`	read	Open a URL in the user\
`open_workspace`	read	Open a user-named local file or directory for LIVE reading, or reuse it. No Git or commit required; includes current uncommitted and ignored files. Original content is read-only.
`opendesign_cli_run`	write	Start an OpenDesign agent run. Returns a runId; poll opendesign_cli_status with it until the run finishes.
`opendesign_cli_status`	write	Poll the status of an OpenDesign agent run by its runId.
`opendesign_export_html`	read	Fetch an OpenDesign project
`opendesign_export_pdf`	read	Export an OpenDesign project deck to PDF. Returns the export result (e.g. a download URL/path).
`opendesign_get_design`	read	Get a single live design artifact by id. Requires both the design id and its projectId.
`opendesign_get_project`	read	Get a single OpenDesign project by id.
`opendesign_health`	read	Probe the OpenDesign instance (GET /api/health, no auth). Use to confirm connectivity before other calls.
`opendesign_list_designs`	read	List the live design artifacts in an OpenDesign project. Requires the projectId, which scopes the query.
`opendesign_list_projects`	read	List the projects in the OpenDesign workspace.
`person_verification`	read	Ask the person, right now, for a sign-in step only they can give, and wait here for it. Exactly three kinds, and the platform refuses any other:
`publish_workspace_file`	write	Store ONE local file on the platform in a single call, and return its
`read_file`	read	Read the contents of a file. Returns the text content.
`read_run_logs`	write	Read a run\
`refresh_workspace`	read	Explicitly refresh from the source’s current committed version. Refuses modified, untracked or ignored files and local commits. Failed preparation preserves the existing workspace. Returns the replacement workspace id.
`relation_add`	write	Record that <subject> —<relation>→ <object>, with what it rests on and why. It is saved first and applied to the graph after: the answer says state
`relation_adopt`	write	Set aside every live edge of one relation name that was NOT recorded through the relation tools (it has no source), so relations recorded before they were maintained never read as current. Nothing is deleted. Running it again finds nothing. Maintainers only.
`relation_confirm`	read	Decide a PROPOSAL: accept makes it a fact (the proposer stays its author, you are recorded as the one who confirmed it), decline retires it with your reason. Maintainers only.
`relation_copies`	read	For many subjects in one call: the latest receipt of each published copy of their relations (relation_copy_record) — where it lives, which version and generation it was made from, the hash of its text, when and by whom — and, as a plain fact, whether the relations have moved since:
`relation_copy_record`	read	Record the RECEIPT of a copy of a subject
`relation_dependents`	read	The same relations read from the other end: for each object, the subjects that assert <relation> to it (\
`relation_history`	write	Every relation operation accepted for a subject, in order: who, at what standing, why, what it changed and the version it produced. Read from the platform
`relation_list`	read	The relations of many subjects in one call. Per subject and per kind of relation: coverage (
`relation_rebuild`	write	Put the graph back in step with the platform
`relation_remove`	destructive	Retire assertions — nothing is deleted, the history keeps who and why. Say WHICH in one of three ways:

Trust audit

BLOCKgrade F · trust 42/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	FAIL
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (5 observation(s))
Network: declared (11 observation(s))
Shell: declared (13 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHCode injection · code.deserialize · CWE-78, CWE-94, CWE-95

src/code-scan.ts:229

pattern: 'yaml.load(...)',

Why it matters. deserialises untrusted bytes into live objects

Fix. use json or yaml.safe_load