Yun Research

Research is legwork you delegate, not thinking you outsource: a background agent does the reading, you keep working, and you get back a cited document to react to.

Steps

1. Recall first. Follow skills/_shared/memory-convention.md to recall prior research on this question, scoped to the target project. If an existing digest already answers it, surface that and stop — no agent needed.

2. Spin up a background agent to do the reading, so the user is not blocked. Hand it the question and this brief:

   • Work from primary sources — official docs, source code, specs, first-party APIs — never a secondary write-up of them. Follow every claim back to the source that owns it.
   • Write the findings to a single Markdown file in the repo, each claim cited to its source. Save it where the repo already keeps such notes; match the existing convention, and if there is none, pick somewhere sensible and say where.

   Done when the agent has returned and the file exists with a source on every claim.

3. Persist the digest. Once the file lands, follow skills/_shared/memory-convention.md to persist a short digest — the question and its findings in a few lines — scoped to the target project. The full file stays in the repo; the digest is what downstream skills recall without re-reading it.

4. Hand it back. Surface the file to the user. It is something to react to — the input that feeds grilling, planning, or a handoff — not a final answer.

Attribution

Adapted for this workspace from Matt Pocock's research (github.com/mattpocock/skills, MIT). Changes: renamed to the yun-<verb>-<noun> harness convention; added a recall-first step and a project-scoped digest via skills/_shared/memory-convention.md (the full cited file still lives in the repo); dropped the OpenAI agent wiring.