log-session

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Log session

Offline study becomes a log entry and node touches. Status moves only on gradable evidence.

## Preconditions

Confirm `lstack.yaml` exists in the working directory or a parent. If not, say this isn't a vault and suggest `/setup-vault`. Read `MISSION.md`, `kb/today.md`, and the newest file in `log/`. Open with one line on where the user is.

## Argument

The text after the skill name in the user's message is free text describing what was studied, when, and for how long, optionally with pasted evidence (solved problems, a written explanation, an image). If there is none, ask what they studied and whether they have anything to show.

## Steps

1. Map the description to nodes by id, title, or `## Terms`. Ask if a mention does not resolve. A topic with no node is proposed as a new node, `status: new`, marked `agent-proposed` unless the user names a source.
2. **From the description alone**, propose:
   - `sessions`: append the study date to each node touched;
   - `last_studied: <old> → <date>`;
   - `confidence` only if the user stated a number;
   - new `sources` entries the user named;
   - aha lines under `## Aha`, notes the user dictated under the right section;
   - `status: new → learning` for a node that was `new`.
   Never any other `status` change, and never `last_tested`, `last_result`, `next_review`, or `passes_in_a_row` from a description.
3. **If evidence is present and gradable**, grade it against the node's Summary, Cards, and sources. Show the grading: each item right or wrong with one line of why, then the ratio `n/m`. Treat it as a test per the scoring block. Say whether it is delayed and therefore counts.
4. **After a description-only log**, offer a two-minute check now: three questions on the node, confidence before each reveal, scored per the block below. Accept a no without argument.
5. Build a log entry `log/<date>-<n>.md` with `mode: offline`, `skills: [log-session]`, the nodes, minutes if stated, `## Covered` from the description, `## Results` from any grading, `## Aha`, `## Changes written`, `## Open threads`. Every heading appears even if empty.
6. Show one numbered batch preview of every node edit and the log file. Wait for yes. Write. Run `node .lstack/lstack.mjs build`, then `node .lstack/lstack.mjs lint`. Report one line. If `lstack.yaml` says `node_available: false`, rebuild `kb/index.md` and `kb/today.md` by hand and say so.

## Scoring for graded evidence (from `lstack.yaml` `scheduling` and `.lstack/node-schema.md`)

- Pass when `n/m >= pass_threshold`. Delayed when `test date - max(last_studied, last_tested) >= delayed_min_days`, using the dates on record before this log. Only a delayed test moves `status`; otherwise write `last_tested`, `last_result`, `sessions` and say "recorded, but this doesn't count toward status".
- Delayed PASS: `passes_in_a_row += 1`; `next_review = date + ladder_days[min(passes_in_a_row, len - 1)]`; `learning → reviewing`; `reviewing → mastered` once `passes_in_a_row >= mastered_after_passes`.
- Delayed FAIL: `passes_in_a_row = 0`; `next_review = date + ladder_days[0]`; `status` drops one step.

## Preview format

```
### Proposed write 1 of 2: kb/vector-spaces/basis.md
**Frontmatter**
- last_studied: 2026-09-08 → 2026-09-14
- sessions: [2026-09-08] → [2026-09-08, 2026-09-14]
**Body**
+ - 2026-09-14: extending an independent list is the same move as reducing a spanning one, run backwards

### Proposed write 2 of 2: log/2026-09-14-1.md (new file)
<full contents>
```

Reply **yes** to write all, **no** to skip all, or name what to change.

## Rules

- Never rewrite an existing log file.
- Never write into `sources/`, `notes/`, or `problems/attempts/`. Never edit generated files.
- No hooks, no sub-agents, no argument substitution.

From sscw32/lstack, MIT.

More from sscw32/lstack