com.deerdawn/deerdawn

Safely configure or repair DeerDawn setup for a supported local surface. Normally called automatically on first authenticated startup via complete_setup — only call this manually if auto-setup failed or you need to repair a specific surface. Creates a backup before writing config files and can restore the latest backup.

Archive a dead or finished project so it stops cluttering list_projects and never auto-resolves as active. Preserves history; pass unarchive:true to restore. Refuses to archive an umbrella that still has live subprojects.

Clean up task context at handoff/finish time. Marks finished task as completed, clears open-task command history, and returns unresolved questions that still need answers.

Run DeerDawn setup in one call: check auth, apply local surface setup, and run import_local_context preview + confirmed import when local context files are found.

Confirm a stored claim still holds after you actually re-verified it (ran the tests, checked the code/state). Stamps it "verified" so the NEXT session trusts it instead of re-opening the code. Targets a recent decision (matched by a phrase from its summary) or the current status (pass "status").

Create a new DeerDawn project. Use this when the user starts working on something new that has no existing project. Pass parent_project_id to create it as a subproject (a workstream under an umbrella project) — subprojects keep their own current task and board.

Search DeerDawn durable path memory across hot paths, entrypoints, surfaces, and debugging notes. Use this before repo search when the file might already be known.

Check DeerDawn authentication status. Call this at the START of every new session before using other tools. If not authenticated, returns a sign-in URL — show it to the user immediately so they can connect.

Get project context (also available as read_brief). Call this when the user asks about a specific project's status, goals, tech stack, open questions, or recent decisions. By default returns only task+status (~30 tokens). Pass sections=[...] to get targeted data, or sections=["all"] for full context. Use start_session at session start instead of this.

Return the most relevant next context to fetch or use for the current task/session, including open questions and pending decisions ranked for relevance and token efficiency.

Return the durable project map for a project: hot paths, entrypoints, named surfaces, and recent debugging notes. Use this before repo-wide search when you need known file-level orientation.

Check DeerDawn local setup status for authentication, global instruction files, and MCP-capable tool configs. Use this when an agent needs to know how well DeerDawn is connected on this machine.

Fetch one shared skill in full (including markdown body) by slug. Call after list_skills when a skill is relevant to the current task.

Query typed workspace knowledge, tools, environments, capabilities, and workflows for drill-down after the overview.

Workspace overview. mode=startup for compact session orientation, mode=overview when user asks "what am I working on?" or wants a full picture of all projects.

Import context from CLAUDE.md, AGENTS.md, or .cursorrules in the current project directory. First call (no confirmed arg) is FREE — no quota used, returns a preview only. Always preview first, show the user, then call again with confirmed=true to actually import. Use proactively on first connection to any project that has these files but no DeerDawn context yet.

Set up a shared skill from a GitHub repo. Paste a repo, folder, or SKILL.md URL — DeerDawn fetches the SKILL.md, reads its frontmatter, saves it as a shared skill, and syncs it to disk for every connected tool (Claude Code, Cursor, Codex).

Read the Deerdawn context knowledge web for a project, session, or specific graph node.

Get a compact index of all DeerDawn projects and available context sections (~30-50 tokens). Call this at the start of a session to see what's available, then call get_context with specific sections for what you need.

List all DeerDawn projects for this workspace. Call this when the user asks what projects they have, what they're working on, or wants an overview of their work. Also use to find the right project_id before calling get_context or update_context.

List available shared skills (name + description, no body). Use on web tools (ChatGPT, Claude.ai) that cannot read skill files from disk.

Manage the active project's task board — a scrum-style board where every task is a card that moves across phase columns: Backlog → Blocked → In Progress → Integration → Testing → Done. Actions: add (new cards, default to Backlog or a named phase), move (re-column a card to a different phase), complete (shorthand for moving to Done), remove (delete a mistaken card), list (show the board grouped by column). This is the single task tracker for the project.

Mark a knowledge or decision entity as resolved so it stops appearing in normal context loads. Resolved entries are still retrievable via get_workspace_entities({ status: "resolved" }). Use when a decision has been implemented, a known issue has been fixed, or a knowledge entry is no longer relevant.

Read your project brief — what you shipped, decided, what's open, and the landmines — the same brief your AI reads at session start. Alias of get_context; returns the 4-slot brief (SHIPPED / DECIDED / OPEN / LANDMINE) with per-item provenance.

Record a durable debugging finding for a project, optionally tied to specific paths. Use this after diagnosing a bug or tracing a non-obvious handoff.

Record durable file or folder hot paths for a project after you discover them. Use this when repo search reveals a path future agents should know up front.

Rename a project (umbrella or subproject). Use to tidy inconsistent, unclear, or shouty project names — the change is immediate and flows into every future brief. Junk names (file paths, "Untitled") are rejected.

Search across ALL projects for decisions, goals, tasks, open questions, pending decisions, or any context matching a query. Examples: "decisions about authentication", "projects using PostgreSQL", "what's blocking the billing work", "open questions about the API", "what decisions do I need to make". Use this instead of get_context when you don't know which project holds the information. If you have not called start_session yet this session, call it first.

Re-parent an existing project: nest it under an umbrella with parent_project_id, or promote it to top-level by omitting parent_project_id. Use to gather related projects under one umbrella after the fact. One level deep; the umbrella must be top-level and the moved project must have no subprojects of its own.

Required first call each session — reads your brief for this project. Auto-detects project from cwd. Returns active context + compact project map (~100-180 tokens). No arguments needed for the common case. (To re-read the brief mid-session use read_brief; to update it use update_brief.)

Set which subproject is active for THIS codebase, so start_session, update_context, and manage_todos target it by default. Remembered per workspace until you switch again. Use when several subprojects share one repo and you want to work in a specific one.

Flush all enabled shared skills to disk for every connected tool (Claude Code, Cursor, Codex). Run after editing a skill so the on-disk copies refresh.

Update your brief — record a decision, an open question, a landmine, or shipped work. Alias of update_context; summary = what changed + why (1-2 sentences). Call immediately when triggered, not batched.

Push a context update (also available as update_brief). summary = what changed + why (1-2 sentences). Call immediately when triggered, not batched. Also stores runnable commands (deploy, test, start) via the commands param.

Create or update typed workspace entities and relationships. Use this for durable knowledge, tools, environments, capabilities, or workflows that should persist beyond a single project note.