council-retro-mini

council-retro-mini

Per-sprint mini retrospective skill for the henkaten-council governance plugin. Runs automatically inline at Step 1H of /henkaten-council:council-autorun after every sprint evaluation completes. This skill is the full implementation of the Step 1H stub that shipped in sprint 6.

Prerequisite: .council/ must already exist in the target project (run /henkaten-council:council-kickoff first).

---

Purpose and Cadence

council-retro-mini provides a lightweight, immediate retrospective capture after every sprint. The goal is to capture learning while context is fresh — NOT to prescribe process changes. Standard-work proposals belong to the pdca mode (/henkaten-council:council-retro) and must not appear here.

Attribute	Value
Cadence	Per-sprint, automatic
Trigger	Step 1H of council-autorun (after every sprint result)
Mode	mini (capture-only)
Time budget	≤30s wall-clock
User input required	No — fully automatic
Standard-work proposals	NO. This mode MUST NOT include any standard-work proposals.
Output path	.council/retrospectives/sprint-{NN}-mini.md

---

Time Budget

This skill must complete within 30s (30 seconds wall-clock). If evidence gathering is incomplete within the time budget, note it in the coverage field and defer analysis to the next pdca cycle. Do not exceed the time limit by attempting comprehensive analysis — concise and timely beats thorough and slow.

---

Mode: mini — Capture-Only

This skill dispatches the Retrospective agent (see agents/retrospective.md) in mini mode. The mini mode contract is:

• Learning Points — what worked, what was harder than expected, what surprised the team (sprint scope only, not cross-sprint synthesis)
• Pattern Observations — early signals of recurring patterns; note that ≥2 sprints are needed to confirm a pattern — label as inferred, not observed

No standard-work proposals in mini mode. This is a capture-only cadence. The Retrospective agent MUST NOT include proposals, kaizen recommendations, or process-change suggestions in mini mode output. Observations that suggest a process change are noted in Learning Points as observations only; the proposal is deferred to the next pdca cycle.

---

Step M0 — Load Sprint Context

Read:

• .harness/sprint-state.json — sprint number and result (PASS / PARTIAL / FAIL)
• .harness/contracts/sprint-{NN}.md — scope and success criteria for this sprint
• .harness/evals/sprint-{NN}-r{R}.md (most recent) — evaluator verdict and evidence

These reads must complete within the 30s budget. If a file is unavailable, note it in coverage and continue with partial data.

---

Step M1 — Dispatch Retrospective Agent (mini mode)

Dispatch agents/retrospective.md as a fork-context subagent via Task in mini mode. Pass:

• Sprint number
• Sprint result
• Path to the most recent eval report
• File paths read in Step M0

Time cap: the subagent's output must return within the remaining time budget (≤25s is the target — 5s overhead for Step M2 write). If the subagent exceeds the budget, interrupt and use whatever partial output was returned.

The subagent is capture-only: standard-work proposals are forbidden in mini mode. If the subagent output contains any proposal for standard-work changes, classify it as a constraint violation and strip the offending section before writing the output file.

---

Step M2 — Write Mini Retrospective File

The Orchestrator (not the subagent) writes the output file to:

.council/retrospectives/sprint-{NN}-mini.md

Use templates/retrospective-mini.md as the scaffold.

Required File Contents

1. Sprint metadata — sprint number, title (from contract), date (UTC ISO 8601), result (PASS / PARTIAL / FAIL)
2. Learning Points — concise bulleted list (3–7 items max to fit the time budget)
3. Pattern Observations — early signal notes; labeled inferred if based on fewer than 2 sprints
4. Henkaten records closed this sprint — list of HK-NNNN records resolved in this sprint's scope (read from .council/henka-register.jsonl)
5. Yokoten block — populated for each closed Henkaten record (see Yokoten section below)

---

Yokoten Propagation

When the mini retrospective closes (resolves) a Henkaten record, the Retrospective agent populates the yokoten block of that record:

yokoten:
  applicable_to_subsequent_sprints: [NN+1, NN+2]   # or ["all"] for universal lessons
  adaptation_notes: >
    {Brief starting point for the adaptation — what the next sprint should
    be aware of or do differently based on this resolved issue.}

applicable_to_subsequent_sprints — list of future sprint numbers that should receive this learning as an adaptation prompt. Use ["all"] for universal lessons. The ratify-once shortcut (v2.1 A9) applies when the value is ["all"] or contains ≥3 sprint numbers.

adaptation_notes — the human-readable starting point for the adaptation. The user (not the agent) drafts the actual adaptation text during Step 1A.5 of subsequent sprints. The agent provides a concise starting point only.

Council-autorun Step 1A.5 (Yokoten Review) reads all henka-register.jsonl records with a populated yokoten block and surfaces adaptation prompts at the start of subsequent sprints. The council-retro-mini skill is the primary writer of yokoten blocks for per-sprint Henkaten closures.

---

What This Skill Does NOT Do

• No standard-work proposals in mini mode. The capture-only rule is absolute. Any finding that suggests a process change is recorded as an observation for the next pdca cycle — not as a proposal here.
• Does NOT invoke the nemawashi walkthrough. Standard-work changes require the PDCA cycle via /henkaten-council:council-retro and Level 5 approval.
• Does NOT perform cross-sprint trend synthesis. That belongs to pdca mode.
• Does NOT require user input. Runs inline and automatically.
• Does NOT persist session context. Context compaction (Step 1G) runs separately in council-autorun.

---

Cross-References

Dependency	Purpose
agents/retrospective.md	Retrospective agent dispatched in mini mode
templates/retrospective-mini.md	Scaffold for the output .council/retrospectives/sprint-{NN}-mini.md file
skills/council-autorun/SKILL.md Step 1H	Caller of this skill; defines the per-sprint stub behavior this skill replaces
skills/council-autorun/SKILL.md Step 1A.5	Yokoten review step that reads the yokoten blocks written here
skills/council-retro/SKILL.md	PDCA retrospective — the cadence where standard-work proposals ARE permitted
.council/henka-register.jsonl	Source for closed Henkaten records and destination for yokoten block writes (via scripts/append-henka.py)
scripts/append-henka.py	Append-only writer used to update yokoten blocks in henka-register.jsonl

---

Output Contract

The output file sprint-{NN}-mini.md is a governance artifact. It:

• Is written once per sprint (idempotent: if the file already exists for sprint NN, log a warning and skip — do not overwrite without explicit user approval)
• Is NOT append-only (it is a single file, not a .jsonl log)
• May be reviewed by the user after the sprint without any action required
• Does NOT block the next sprint — the autorun loop proceeds to Step 1I immediately after this file is written