Sprint Retro (PM)

Gathers the data and sets the table for the retro. Does no evaluating, judging, or concluding (that is the team's job, in the meeting).

Sources to read (this sprint's scope)

• Connections and patterns first via graphify (structure): treat the god-nodes (highly connected core concepts) and surprising connections (unexpected cross-links) in graphify-out/GRAPH_REPORT.md as retro-material candidates. For delayed or blocked items, use graphify explain "<concept>" and graphify path "<A>" "<B>" to trace structurally what the item depends on or connects to that caused the blockage (e.g., "login delay" is connected to the "switch-to-Google decision" and "store review"). Do not read everything wholesale; read only the sources the graph points to.
• Jira backlog: done/in-progress/blockers + plan vs. actual (completion rate, carry-over) + (if recorded) estimated vs. actual time. Progress checklists are not in the lore vault (they live in each team member's dev repo), so treat Jira as the source of truth for progress and time.
• wiki/ (concepts, decision history): decisions and reversals made during the sprint (only what graphify points to)
• conflicts/: conflicts that occurred and were resolved

Retro draft structure

1. Plan vs. actual — completion rate, carry-over (as far as the data goes).
2. Estimated vs. actual time feedback — if Jira has estimate/actual times, lay them side by side per item to surface estimation accuracy (e.g., "login estimated 2d / actual 5d — 2.5x over"). If there is no data, mark "no time data" and gently remind the team to record estimates and actuals in Jira. This is for calibrating estimates, not for evaluation.
3. What went well / what got stuck — based on blocker and conflict patterns.
4. Decision retro — reasons behind reversed decisions (from decisions).
5. Discussion questions — phrase improvements as questions (never force conclusions).

Provide the minutes form + polish

• Along with the retro draft, present an empty meeting-minutes form in Slack so a human can easily fill it into raw/:

  # Sprint Retro — <sprint>/<date>
  ## Attendees
  ## What went well
  ## What got stuck / causes
  ## Estimated vs. actual (estimate calibration)
  ## Action items (owner / due date)

• Once a human fills in this form and records it in raw/ (e.g., raw/meetings/retro_<date>.md), polish that meeting note on invocation — typo fixes, structure cleanup, and item ordering only. No content invention, no added conclusions (blanks stay blank). The polished result is also applied only after human confirmation.

Rules (invariants)

• No evaluation, no blame, no judgment. ("A was slow" ✕ → "task X delayed N days, cause needs discussion" ○)
• Only what has evidence in the lore vault. Never invent events or numbers. If there is no data, mark "no data".
• Retro outcomes (actions) are recorded by a human as meeting notes in raw/ → fed back through the normal pipeline. This skill does not finalize decisions or actions.
• This channel = this project only. No mentioning or referencing other projects.

Output

• Post the retro draft to Slack (sections 1–5 + evidence sources) + the empty minutes form. No automatic minutes writing (scope ends at providing the form and polishing).