원클릭으로
tsq-end-session-memory
Extract and save project memory (facts, decisions, conventions) before a TaskSquad session closes
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Extract and save project memory (facts, decisions, conventions) before a TaskSquad session closes
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | tsq-end-session-memory |
| description | Extract and save project memory (facts, decisions, conventions) before a TaskSquad session closes |
This session is closing. Before it ends, reflect on what happened and save any memory-worthy facts about the project — not reusable procedures (that's /tsq-end-session-learning), but facts: things a teammate or another agent working on this project would benefit from knowing.
Review this session against each of the 5 memory categories below. For each one, decide: nothing (the common case — most sessions produce a memory-worthy fact in zero or one category, rarely more), a local entry, or a global entry.
Every entry you decide to save also needs a scope:
When in doubt, ask: "would this be useful to an agent running on a teammate's laptop?" Yes → global. No, this is specific to this checkout or machine → local.
/tsq-end-session-learning, not hereGood (architecture / global):
title: Session transcripts stored in R2, not D1
category: architecture
tags: storage, transcripts
Full session transcripts are uploaded encrypted to R2 rather than stored in D1,
to keep D1 rows small. See packages/worker/src/routes/daemon.ts.
Good (structure / global):
title: Local install layout is shared across skills/agents/commands
category: structure
tags: cli, conventions
Skills, sub-agents, and commands all install to <work_dir>/.tsq/<kind>/<name>/.
A new install-type should follow the same layout rather than inventing a new one.
Good (personal / local — machine-specific, not useful to teammates):
title: Local dev uses a forked tsq binary
category: personal
tags: local-setup
This checkout runs a personal fork of the CLI at ~/dev/tsq-fork for testing —
don't assume the published binary when debugging CLI behavior here.
Bad (too trivial, don't save):
Ran `npm install` to install dependencies.
Bad (this is a skill, not memory — push it with tsq skill instead):
To apply a D1 migration remotely, run `wrangler d1 migrations apply --remote`.
Use your own Write tool directly — no CLI round-trip needed, since local memory lives inside your own work_dir and never touches the network. Write to:
.tsq/memory/<category>/<YYYY-MM-DD>-<kebab-slug-of-title>.md
With frontmatter:
---
title: <short title>
category: <one of the 5 categories>
date: <YYYY-MM-DD>
tags: tag-one, tag-two
---
<markdown content — the fact itself, with enough context to be useful later>
Write the content to a temp file, then push it with tsq memory push:
# 1. Write memory content to a temp file
cat > /tmp/tsq-memory-entry.md << 'EOF'
---
title: <short title>
category: <one of the 5 categories>
tags: tag-one, tag-two
---
<markdown content>
EOF
# 2. Push to the daemon
tsq memory push \
--scope global \
--category <one of: personal | preferences | structure | architecture | events> \
--title "<short title>" \
--tags "tag-one,tag-two" \
--file /tmp/tsq-memory-entry.md
A HTTP 200: {"id":"...","title":"..."} response means it was saved successfully.
If you need to check whether something is already known before saving a duplicate, search both scopes at once:
tsq memory search "<query>"
Results are tagged [local] or [global] so you know where each fact came from.
After saving all memory-worthy items (or if there are none), respond with a brief summary: