一键在 Manus 中运行任何 Skill

session-retro

星标2

分支0

更新时间2026年7月1日 21:05

This skill should be used by the orchestrator at the end of pipeline commands (team, quick, design, debug, gc, address-comments) after /kon:summarize, to propose session learnings for public or repo memory with human confirmation before write.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

dentiny

dentiny/kon

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Session Retro

Owner: orchestrator (no Task subagents) Consumers: /kon:team, /kon:quick, /kon:design, /kon:debug, /kon:gc, /kon:address-comments. Optional manual re-run: /kon:retro.

Core principles (always): follow skills/core-principles — propose only evidence-backed lessons; don't invent retro insights.

Save paths

Saving is always:

Mid-session — Mio/Yui ## Memory propose → confirm → write
End of pipeline — retro (below) → confirm → write

To browse stored entries, read the indexes directly:

cat ~/.kon/public/memory/MEMORY.md
cat ~/.kon/projects/<repo-name>/memory/MEMORY.md

When to run

Default session close for pipeline commands (after work + Mio approve + /kon:summarize):

summarize → retro → user /kon:finish or dashboard ✓

Skip retro for read-only commands: /kon:ask, /kon:review, /kon:review-pr, /kon:research, /kon:describe-issue, /kon:todo.

User may say skip retro to close without proposing.

Candidate sources (≤ 5)

Scan conversation + session artefacts:

Explicit user prefs / feedback ("always integration tests", "shorter reviews")
Conventions agreed this session (commit style, test naming)
Lessons / pitfalls ("lib X differs on macOS")
Skipped ## Memory propose sections from Mio/Yui this session (re-offer)
Plan ## Honoured memory or debug notes worth persisting

Exclude: one-off task facts, ticket numbers, secrets, .env values, API keys.

Flow

Collect ≤ 5 candidates — rank by reuse signal; drop lowest if > 5.
Zero candidates → print "Nothing to propose for memory." and exit retro.
One at a time — for each candidate:
- Print Candidate i/N: quote + inferred type, name, description, suggested scope (public or repo).
- Ask: Save to public / Save to repo / Edit / Skip / Done with retro.
- On save/edit → run skills/memory-propose-confirm write steps.
Summary — list saved entries with scope, or "Retro complete — nothing saved."

Scope heuristic (user may override)

Signal	Suggest
"I always…", language, review/commit habits	public
"In this repo…", file paths, local tooling	repo
Unclear	public for `user`/`feedback`; repo for `project`

Path A vs B

Path A — same conversation: use chat + sessions/<id>/summary.md + .kon/plan-*.md.
Path B — standalone /kon:retro when conversation context is thin: ask the user what to save, then run step 3.

Orchestrator rules

Narration: 🌸 Ui per skills/narration.
One candidate at a time — no multi-select batch.
Do not infer beyond evidence — only what user said or session artefacts show.
Do not delegate — orchestrator runs retro; no agents.
Memory ≠ waiver — saved conventions inform agents; they do not lower Mio's bar.

Instant propose vs retro

Both may run in one session. Retro still runs after summarize even if mid-session proposes occurred. Skip re-proposing entries already saved this session (user confirmed write).

同仓库更多 Skills

同仓库

design-debate

dentiny/kon

Multi-agent design argument — Azusa challenges Mugi's plan, Mugi revises. Used by /kon:design after exploration and initial plan draft.

2026-07-062

kon

dentiny/kon

Multi-agent software engineering workflow. Use this skill whenever the user types a /kon: command — including /kon:team, /kon:design, /kon:quick, /kon:debug, /kon:ask, /kon:review, /kon:review-pr, /kon:hunt, /kon:gc, /kon:research, /kon:describe-issue, /kon:begin, /kon:finish, /kon:summarize, /kon:retro, /kon:address-comments, /kon:todo, /kon:understand-codebase. Orchestrates specialized agents: Azusa (explorer), Jun (researcher), Mugi (planner), Yui (implementer), Mio (reviewer), Sawako (cleaner), Nodoka (summarizer).

2026-07-062

memory-loading

dentiny/kon

This skill should be used by all kon agents at startup, before beginning work, to load relevant memory from public and repo indexes with relevance-based ordering and a 10-entry cap. Consumers: Azusa, Mugi, Mio, Jun, and any future memory-reading agent.

2026-07-062

repo-skill

dentiny/kon

Convention for per-repo skill directories at ~/.kon/projects/<repo-name>/skills/<name>/SKILL.md. Each named subdirectory is one skill. Agents load all SKILL.md files as the first step before memory entries.

2026-07-062

ask-dont-guess

dentiny/kon

When requirements, code behavior, root cause, or review scope is unclear — ask the user before proceeding. Never invent facts, assumptions, or fixes. Applies to every kon stage and every agent.

2026-07-062

interactive-session

dentiny/kon

Route plain-text user messages while a /kon:begin session is open. Reuse the same session JSON — never init a second session for sub-turns.

2026-07-062

name	session-retro
description	This skill should be used by the orchestrator at the end of pipeline commands (team, quick, design, debug, gc, address-comments) after /kon:summarize, to propose session learnings for public or repo memory with human confirmation before write.

Session Retro

Owner: orchestrator (no Task subagents) Consumers: /kon:team, /kon:quick, /kon:design, /kon:debug, /kon:gc, /kon:address-comments. Optional manual re-run: /kon:retro.

Core principles (always): follow skills/core-principles — propose only evidence-backed lessons; don't invent retro insights.

Save paths

Saving is always:

Mid-session — Mio/Yui ## Memory propose → confirm → write
End of pipeline — retro (below) → confirm → write

To browse stored entries, read the indexes directly:

cat ~/.kon/public/memory/MEMORY.md
cat ~/.kon/projects/<repo-name>/memory/MEMORY.md

When to run

Default session close for pipeline commands (after work + Mio approve + /kon:summarize):

summarize → retro → user /kon:finish or dashboard ✓

Skip retro for read-only commands: /kon:ask, /kon:review, /kon:review-pr, /kon:research, /kon:describe-issue, /kon:todo.

User may say skip retro to close without proposing.

Candidate sources (≤ 5)

Scan conversation + session artefacts:

Explicit user prefs / feedback ("always integration tests", "shorter reviews")
Conventions agreed this session (commit style, test naming)
Lessons / pitfalls ("lib X differs on macOS")
Skipped ## Memory propose sections from Mio/Yui this session (re-offer)
Plan ## Honoured memory or debug notes worth persisting

Exclude: one-off task facts, ticket numbers, secrets, .env values, API keys.

Flow

Collect ≤ 5 candidates — rank by reuse signal; drop lowest if > 5.
Zero candidates → print "Nothing to propose for memory." and exit retro.
One at a time — for each candidate:
- Print Candidate i/N: quote + inferred type, name, description, suggested scope (public or repo).
- Ask: Save to public / Save to repo / Edit / Skip / Done with retro.
- On save/edit → run skills/memory-propose-confirm write steps.
Summary — list saved entries with scope, or "Retro complete — nothing saved."

Scope heuristic (user may override)

Signal	Suggest
"I always…", language, review/commit habits	public
"In this repo…", file paths, local tooling	repo
Unclear	public for `user`/`feedback`; repo for `project`

Path A vs B

Path A — same conversation: use chat + sessions/<id>/summary.md + .kon/plan-*.md.
Path B — standalone /kon:retro when conversation context is thin: ask the user what to save, then run step 3.

Orchestrator rules

Narration: 🌸 Ui per skills/narration.
One candidate at a time — no multi-select batch.
Do not infer beyond evidence — only what user said or session artefacts show.
Do not delegate — orchestrator runs retro; no agents.
Memory ≠ waiver — saved conventions inform agents; they do not lower Mio's bar.

Instant propose vs retro

Both may run in one session. Retro still runs after summarize even if mid-session proposes occurred. Skip re-proposing entries already saved this session (user confirmed write).