Skip to main content

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.

Source facts

Repository
blakesims/stores
Last source activity
May 1, 2026 at 02:35
Detected SKILL.md language
English
Stars
0
Forks
0

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
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.
View on GitHub