## Context

Work that needs a plan runs through Context (https://app.onecontext.me). The
Agent guide (https://agents.onecontext.me/) is the standing instruction; the
`context` skill installed in this repo's skills directory wraps the tools.
In short:

- Call `setup` once per session before the first write; if it reports the
  Guide or a skill as `update` / `install`, refresh your copy first.
- Brainstorm with `grill-me` + `wayfinder` (Matt Pocock, MIT); never a
  from-scratch questionnaire. Facts are fetched, decisions are asked.
- Chart exactly one Context map (Epic + Issues + `blocked_by` edges) plus a
  spec; raise one `request_review` for both together; wait on `get_changes`.
- Work Issues one at a time (or in parallel by dependency graph, one
  worktree each). Every ticket attaches a mini-spec, acceptance tests and a
  release doc; the map gets an overall release doc.
- Claim completion with `claim_issue` (evidence) and get it checked with
  `verify_issue` from a different principal — a claim is not a check. Never
  mark a human gate done yourself.
- Skills ship a benchmark. Code specs follow the `context` skill's
  `references/spec-template.md`.
- Recurring workflows install `daily-brief` first.
- Report `work_stats` on `complete_issues`; key names only, never values.
- Post a handoff comment on the ticket before stopping.