CLI Companion Personas

`ed companion personas`

Lists the lenses that can answer you, and how each one thinks. A persona is not a tone: it is a spec file that sets which sources are read, how much your own account of yourself counts against independent records, which passes run, what shape the answer takes and how sure the lens must be before it says anything. The voice is the last line of the file.

Usage:

ed companion personas [--json] [--endpoint <url>]

The four that ship:

Lens	How it reads you
`analyst`	Evidence first. Weights records of what you did above your account of it, and says when the record is thin.
`friend`	Long window, high salience weighting, warm because it recalls the specific thing you said, not because it agrees.
`coach`	Last thirty days, commitment focused, ends with exactly one next action.
`skeptic`	Argues the other side. Runs a counterfactual pass and a second inverted retrieval, and abstains unless the record carries the push.

--json shape: an array of {id, label, pipeline, output, abstainBelow, maxWords, selfReportWeight, observationWeight, k, windowDays}.

Adding a lens is data, never code: drop a YAML spec and a voice prompt into the directory named by PERSONA_DIR on the backend. Files are <id>.yaml and the optional matching <id>.md; custom ids replace built-ins with the same id. Malformed specs and unknown pipeline stages are skipped and written to backend stderr. The returned list is sorted by id.

Pass any of these to ed companion ask with --persona, or ask several at once with ed companion council. Chat also accepts --persona and defaults to friend; ask defaults to analyst.