Tools and actions — Annsa Docs

01The four tools

Annsa gives your agent four MCP tools. annsa.priorities reads the ranked list, or one priority by number (priority_number=N; mine=true shows what this seat holds). annsa.spec reads the spec for a priority. annsa.act does everything that changes something. annsa.ask answers questions and serves the skills (topic=working, spec-skill, decisions, welcome), and checks do_we_have before you file. annsa.help and annsa.surveys are older names for annsa.ask and still answer.

A priority you just filed can be missing from the next list read for a few minutes. Ask for it by number instead.

02Who may run an action

Three groups. Agent means an agent credential may run it. Person means the agent credential is refused, on purpose: these are decisions, and an agent that could make them would be marking its own homework. Partly means an agent may run one form of it. The credential decides, not the client, so Claude Code signed in with OAuth is you and has your rights, and an agent credential you minted does not. who.actor.kind says user or agent.

03Work on a priority

Action	What it does	Who may run it
`claim`	Take a priority before touching code. Creates the spec if there is none. Two seats can't hold one priority without `takeover=true` and a reason. (`build` is the old name.)	Agent
`note`	Add a progress note to the priority; send `pr_url` to put the PR on it.	Agent
`release`	Put the work down without shipping. Needs a reason, and the ruled-out options are kept.	Agent
`handoff`	Pass the work to someone with what's true and the next step.	Agent
`ship`	Say the work is done, backed by the receipts recorded. Never deploys. `learned` lands as an idea for the next spec.	Agent
`verify`	Record how one Done When line was checked, with a `where` someone can follow. Optional per line, never blocks a ship.	Agent
`status`	Move a priority's planning stage (open, ready, closed).	Agent, only with workspace planning authority
`retract`	Take back a claim that turned out wrong, such as feedback filed on the wrong priority or a ship that didn't happen. Refused once customers were told.	Person

04Write the spec

Action	What it does	Who may run it
`write`	Write one section of a spec you hold: What to build, Files to touch, Done When. Anything that isn't yours becomes an update with a reason.	Agent
`approve` / `dismiss`	Decide an update someone else suggested. An agent can decide only on a section it may write right now. (`accept` and `decline` are the old names.)	Agent, or person
`correct`	Fix a priority's title, detail or ruled-out entry, with a reason. Category corrections teach the classifier.	Partly: title, detail and ruled_out for an agent; category is a person's
`direction`	Read the direction document, or write a new version of it.	Partly: an agent can read, and its update is a proposal until a person accepts it

05File and tidy

Action	What it does	Who may run it
`submit`	File a row of feedback or a finding. `stand_alone=true` always mints a new priority. Ask `do_we_have` first. (`adopt_priority` is the old name for stand-alone.)	Agent. Its rows count as findings, never as a person
`merge`	Fold priorities into one survivor. The merged numbers are retired and never reissued. Refuses anything shipped.	Agent
`regroup`	Move priorities between groups. Renames buckets only.	Agent
`severity`	Set P0 to P3 on a finding.	Partly: an agent proposes, it counts once a person confirms, and a person's severity always wins
`transcript`	Bring in a call transcript as customer voice.	Person
`import_direction`	Import a strategy document; over MCP it proposes ideas you choose from.	Person

06Decide and direct

Action	What it does	Who may run it
`bet`	`list` shows the bets. `place` moves priorities onto a bet. `create`, `update`, `delete` and `reorder` change the frame.	Partly: an agent can list and place; a person owns the bets
`sign`	Record a belief on a priority: do this, do nothing, park, dismiss.	Person
`park`	Set a priority aside with a reason and a date it comes back.	Person
`unpark`	Put a parked priority back in front of a person. Lifting a signed park needs a reason.	Agent
`assign`	Give a priority to a person or an agent.	Person
`principle`	Record a rule with a scope (priority, area or company).	Partly: an agent can propose; a person writes and activates

07Ask and answer

Action	What it does	Who may run it
`escalate`	Raise a question in Needs You on one of four triggers, with a three-line ask. One open ask per priority.	Agent
`answer`	A person's reply to an escalation.	Person
`ask_back`	A person asks the agent something back; the ask stays open.	Person
`adjudicate`	Settle a Needs You question where the model was unsure, or sure and wrong.	Person

Settled topics aren't asked again. Signed in as yourself, you can answer from your own terminal. Only a "Looks like #NNN" answer merges on its own; every other answer is stored, and the agent reads it on the spec as decision if the priority has a spec.

08Tell customers and cut releases

Action	What it does	Who may run it
`share`	Send the note to the customers behind a shipped priority.	Person
`cut`	Close a release for a period: every shipped priority plus a changelog summary in the rows' own words. Prepares no share-back. Runs over MCP only; there's no cut button in the app.	Agent

Next guideWhat your agent can do