write-insights

---

name: write-insights description: Use when an Exploration cycle is wrapping up and the insights.md learning artifact is needed (mandatory before G2), or when Production has revealed implementation surprises that must be appended (Rule 9 / Learning Loop). Trigger phrases: "write insights", "insights file", "what did we learn", "log insights", "update insights with implementation surprises".

Write Insights (Learning Loop)

insights.md converts the lifecycle from a one-way pipeline into a loop. It records what was tested, what surprised us, and what we now believe — feeding back into hypotheses and risks for the next cycle.

Mandatory before G2 (initial draft). Updated again after G3 with implementation surprises (Rule 9).

When to trigger

• An Exploration cycle is approaching G2 and insights.md does not yet exist
• A prototype, measurement, or interview has produced findings the user wants captured
• Production / validation surfaced surprises that must be appended (Rule 9)
• User explicitly asks: "write insights", "log learning", "update insights with what we found"

Steps

1. Read the template: templates/insights-template.md.
2. Read related artifacts to ground the insights:
   • The active EXP plan (for hypotheses to test against and the ## Vision Alignment section)
   • hypotheses.md if present
   • Any prototype output, measurements, interview notes, or ## Challenger block
   • risks.md (new surprises become new risks)
3. Locate the insights file — typical path: context/plans/<plan-id>/insights.md or alongside the EXP plan file. Read CLAUDE.md or ask; do not guess.
4. Draft the file satisfying the template's Minimum Substance:
   • At least one row in Hypotheses tested with non-empty Result and Verdict (confirmed / refuted / inconclusive)
   • At least one bullet in What surprised us — be specific; "nothing surprised us" is itself an insight, but document why
   • At least one bullet in What we now believe
   • Open questions in What we still do not know
   • One of the four Consequence for the plan checkboxes ticked (proceed / adjusted / killed / fallback to Context)
5. Cross-update dependent artifacts the user agrees to:
   • hypotheses.md — mark hypotheses confirmed / killed
   • risks.md — append new failure modes the Exploration uncovered
6. Confirm path and content with the user before writing.

Rules

• Insights are written during Exploration, not retro-fitted after G2 approval (the template flags retro-fitting as an anti-pattern).
• "Confirmed everything" is a smell. If every hypothesis is confirmed and What surprised us is empty, regenerate with sharper questions to the user before saving.
• For G3 updates, append a new dated section rather than rewriting earlier findings — the audit trail matters.
• Status is draft until the human confirms; only the human fills decided-by: (required at G2).
• If a prior ## Challenger failure mode came true, record that linkage explicitly — it closes the loop the Challenger opened.

Output

An insights.md file (or an updated one) that satisfies the template's Minimum Substance requirements. Status draft until human confirmation.