observation-log
Quickly log something noticed mid-flow into the observations store. Triages first; only files when the fix isn't a T1 you can do right now.
ソース情報
- リポジトリ
- blakesims/stores
- ソースの最終更新活動
- 2026年5月1日 02:35
- 検出された SKILL.md の言語
- 英語
- スター
- 0
- フォーク
- 0
インストール方法
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
ソースファイルを確認
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SKILL.md を表示中
SKILL.md
ソースの指示 · 読み取り専用プレビュー- name
- observation:log
- description
- Quickly log something noticed mid-flow into the observations store. Triages first; only files when the fix isn't a T1 you can do right now.
- requires_stores
- ["observations"]
- default_invoker
- ai_with_human
- user_invocable
- true
# Log an observation
The user noticed something during a working session and wants to capture it.
Aim for **10 seconds**: triage first, then either fix in-session (T1) or file
an entry (T2/T3). Do not investigate.
## Discover the surface
Once per fresh session:
```bash
stores observations --help # verbs
```
The schema tells you what fields `add` accepts. If it changes, your CLI calls
adapt automatically.
## The judgment
Apply this rubric (it lives here, not in the schema, because it's reasoning,
not data):
| Tier | Trigger | Action |
|------|---------|--------|
| **T1** | ≤2 files, ≤50 LOC, no migrations/deps/config | Propose the fix inline. If user approves, edit + commit. **Do not file.** Commit message is the durable record. |
| **T2** | ≤5 files, ≤200 LOC, single subsystem | File the entry; downstream `/observation:triage` will route to a mini-loop. |
| **T3** | Migrations, cross-subsystem, capability change | File the entry; downstream `/observation:triage` will gather the contract. |
| **Can't tell** | Not enough context to triage confidently | File with `--body "needs triage"` so triage classifies later. |
Override: if the user says "just log it," skip the rubric and file.
## Action
```bash
# T2/T3 capture (default path):
stores observations add \
--summary "<1-line description>" \
--source dev \
--priority normal \
--captured-at "$(date -I)" \
--captured-week "$(date +w%V-d%u)" \
[--body "<extended notes>"] \
[--contact-id <n>] \
[--field-name <name>]
# Long descriptions: pipe via stdin
stores observations add --summary - <<< "$(cat <<'TXT'
multi-line summary here
TXT
)"
```
`--summary`, `--source`, `--priority`, `--captured-at`, `--captured-week` are
all required by the schema, but the latter two are mechanical — the
`$(date ...)` defaults above fill them without operator input. No
`--invoker` flag is needed on `add`: none of the fields written here carry
an `actor:` constraint, so the auto-detect default is fine.
The CLI returns the new display_id (e.g. `L042`). Confirm to the user in one
line: `Logged L042 (priority): summary`. Done.
## Rules
- One observation per invocation. No batching.
- Default `--source dev` unless the user says Sentry / dashboard / intake.
- Infer fields from conversation context. Don't ask for what's already said.
- Triage before filing.
- Never call other skills from here.
GitHubで見る