flow

Product discovery · Portable skill

research/

Investigate a question or a competitive landscape from sources and record a dated, sourced note another session can decide from.

npx skills add gaboesquivel/flow-skills -s research

Say it

Natural language is the interface. Any of these, or your own words, reach the skill.

  • “research X”
  • “competitors for X”
  • “research X, just tell me here”

When to use it

A sourced note answering the question, or a competitive brief for "competitors for X": the decision it serves, findings marked observed, claimed, or inferred, limits, and implications, updating a current note instead of duplicating it.

The skill

The authoritative, versioned instruction your agent follows, from skills/research/SKILL.md.

"Research X" answers a question from evidence and leaves a note someone who missed the conversation can decide from; "competitors for X" is the same job as a competitive brief. The request authorizes the note.

What the request asks

  • "Research X" or "competitors for X" in a project: write or update the note, then link it.
  • "Just tell me here", or no known project destination: answer in chat in the same shape.
  • "Should we research…?", a quoted phrase, or a hypothetical: discuss whether research would change the decision; write nothing.

Investigate

Start from the decision the research serves; infer it and say so when unnamed. Read the project context it touches and existing notes first: update a current note on the same question rather than duplicating it, keeping superseded findings readable with their dates. Then look externally, preferring primary sources (official docs, repositories, the product itself).

For a competitive brief, establish the intended user, the job they are doing, and the realistic substitutes: the status quo or a manual workaround, direct products, and open-source tools, host features, APIs, or integrations. Compare only distinctions that could change the choice, in a table when it is a genuine comparison; no generic feature matrix or exhaustive list.

Rules of evidence

  • Every material finding links a direct source, and time-sensitive ones carry the date read.
  • Separate observed (you read or ran it), claimed (a vendor or author says so), and inferred (your reasoning).
  • Never invent a source. A feature you did not find is "not found in <source>", never absent.
  • Implications are proposals. A note never adopts a recommendation, edits PRODUCT.md, or starts setup or delivery; name the skill that would (explore, sync-docs, to-spec) and stop.

The note

Use the project's research convention (Flow Skills: docs/research/, indexed in docs/references.md); without one, write a single docs/research/YYYY-MM-DD-topic.md and no index, taxonomy, or tree. As short as complete:

  • Question and the decision it informs; the date and what was inspected.
  • Findings, each sourced and labeled observed, claimed, or inferred; for a brief, the user, job, substitutes, and distinguishing differences.
  • Limits and unknowns: what was not read or failed to load, and what would change the conclusion.
  • Implications: the strongest alternative and what the findings suggest for the decision.

Write only the note and an existing index, on a non-default branch (the current one, else one named for the research; write-agents Sessions); commit and report the path as a link, opening it where the host supports that.

Jump to, or try a question