Bourdon

Capture a research note, code snippet, or link into Bourdon memory. Saves to the federated clipper store for discovery by all agents.

Call when the conversation settles something the user's other agents should know: a decision, a plan, a status change (shipped, blocked, paused) or a finding. Commit under your own agent_id. Mark anything personal, financial or credential-like visibility: "private". Never commit small talk or a transcript. Write a contribution to the federation under agent_id: durable facts in entities (name + summary), what happened in sessions (date, project_focus, key_actions), short dated facts in notes.

Compile a turn-scoped Codex recognition brief.

Call FIRST, before answering, whenever the user's message references their projects, people, decisions or prior work. Returns a short recognition brief of what the user's other agents already know. `format` picks markdown, xml, compact, brief, json or yaml. `verbose: true` returns the full brief with every delivery variant, trace and diagnostics.

Describe this authenticated member and current meetings authority. Requires explicit library enrollment and member-bound credentials. A descriptor is a policy observation, not a reusable write permit.

Which write produced a given state of an agent's memory: hand it a manifest digest (or a version's digest8) and it returns the caller, session, transcript, and what changed. session_id and transcript_ref are caller-asserted and unverified; the recorded caller identity is not.

Export THIS server's LOCAL agents only, source-attributed for the tray.

Find an entity by name across all agents.

Aggregate everything the federation knows about a project.

Return post-recognition L2 context for the prompt.

Load one shared skill: its `SKILL.md` text as `body`, with provenance. Only skills promoted into the shared library have a body here. The response carries `source_agent`, `promoted_by`, `version` and `content_sha256` beside the text so you can see what you are about to follow. `version` selects an archived version (default: the live one; a retired skill is served only by explicit version). `include_files` adds the reference files (`file`, `bytes` and `text`). For a skill an agent only advertises, the response is its index rows and a message that it must be promoted first; Bourdon never reads an agent's skill folder over MCP.

Ingest a meeting transcript, sync notes, or decision log into meetings memory.

List agent IDs known to this L6 server, plus any peers' agents.

Prior states of one agent's memory, newest first. Each row carries an opaque version_id to pass to rollback_agent_memory and a digest8 that joins to the provenance entry explaining which session produced it.

Return a page of freestanding notes across agents.

Return a page of sessions across agents (or a single agent).

List the skills (step-by-step procedures) this caller may load. Three kinds of row, told apart by `tier`: * `shared` -- promoted into the shared skill library. Carries `name`, `description`, `version`, `content_sha256` and its provenance (`source_agent`, `promoted_by`, `promoted_at`). `get_skill` returns its body. * `advertised` -- a skill an agent keeps locally and lists in its manifest (only with `include_local`, the default). One row per `content_sha256`, naming every agent that has it in `advertised_by`, with `origin` when known and `shared` (`matches` / `diverged`) when a shared skill has the same bytes or name. Not loadable until promoted. * `proposed` -- a skill proposed for the shared library (`propose_skill`) and awaiting a person's approval (only with `include_local`; a trusted caller sees every pending proposal, any other caller its own). `query` filters by a case-insensitive substring of the name or description. Rows follow the caller's access ceiling; advertised rows come only from namespaces the caller may read.

List ongoing multi-session workstreams across agents.

Return immediate recognition and a bounded prompt-context fragment.

Propose a decision topic for multi-agent federation consensus. Strategies: majority, supermajority, unanimity, weighted.

Propose one of your advertised skills for the shared skill library. Bourdon reads the skill from the folder your own skill reader advertised (never a path you pass), checks that it is text-only, and snapshots it into a pending proposal, so editing the folder later does not change what gets approved. A person approves or rejects it at a terminal (`bourdon skills approve`); no tool approves a proposal. `content_sha256` picks one advertised copy when you advertise more than one skill with this name. `source_agent` is honoured for the local operator only; every other caller proposes its own skills. Pending proposals appear in `list_skills` as `tier: proposed`.

Find entries in one agent's L5 that match a topic.

Restore an archived version of an agent's memory. TRUSTED ONLY — the most destructive operation the server exposes: it discards every contribution made after the chosen version, including other agents'. Pass expected_digest when the decision took a while, so a commit that landed meanwhile fails the restore instead of being silently erased. The restore is itself snapshotted first, so a rollback is undoable.

Search across federated agent memory, entities, consensus proposals, and clips. Uses sublinear-scaled TF-IDF cosine similarity vector search.

Trace the memory provenance and evolution of an entity across all agents. Inspects which agent session originated it, which git commit / transcript was referenced, which version archive snapshots recorded it, and which peers also carry it.

Cast a vote on an active multi-agent consensus proposal.