oguzdogmus/mimir / skills/mimir-core-rules
mimir-core-rules
VerifiedHard rules for Mimir, the exam study plugin. Load this before any other Mimir skill runs and before reading, writing, or acting on a course workspace (the learning root, default ~/learning/, or any Mimir .state/ file).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 20.8 KB |
| references/recording.md | markdown | 1.8 KB |
| references/session-close.md | markdown | 1.8 KB |
| references/verified-items.md | markdown | 4.0 KB |
references/verified-items.md
View on GitHub# Verified items: gate, one wait, verifier batch, retry, final gate
Shared by every Mimir skill that shows questions with answer keys (`quiz`, `study`). The calling skill has already generated its items (per item: `topic`, `origin`, and for a real item `source`, `page`, `question`; optional `quote`, `compare`; the question with its marks, a worked answer key and one hint) and knows the course `term`, `course` and `course_dir`. Gate calls: `python3 "<mimir>/scripts/verify.py" gate` with JSON on stdin in a quoted heredoc (`<<'EOF'`).
1. Gate: `verify.py gate` with `{"term": "...", "course": "...", "items": [{"topic": "...", "origin": "...", ...}]}`, without verdicts. Items carry only `topic`, `origin`, `source`, `page`, `question`, `quote`, `compare`, `verdicts` and `dispute`; question text, answer keys and hints stay with you. Every gate call (here and in steps 3–5) sends the full item list in the same order, so each item keeps its number `N`.
- Exit 2 naming `items[N].<field>`: a real item's `source`, `page` or `question` gets the correct citation from `INDEX.md` or the item becomes `practice`; any other field is fixed; rerun. In this step only, if the same field is named twice, remove that item and rerun. After step 2 no item is ever removed: if a fix fails twice, make that item `practice` (no `source`, `page`, `question`, `quote` or `compare`), keep its `verdicts`, and rerun.
- Any other exit 2 (not naming `items[N].<field>`): print its stderr line and stop.
2. If `verify` is empty, that gate is the final one: go to step 5. Otherwise, on the session's first verifier call only (a whole quiz is one session), say: "Checking my answers before you study them, about 30 seconds." Later calls in the same session say nothing. Then dispatch ONE verifier batch using core-rules’ capability-based helper dispatch, passing the resolved `<mimir>/scripts/pdftext.py` path and helper body, with all `verify` items: per item its number `N`, the gate's `label`, the question, the answer key, and its sources as absolute paths (`course_dir` + registry path) with pages, or short excerpts. A real item always gets its `source` as an absolute path with its `page` and `question`; an `ocr` one also gets the gate's `original`. An item with a `quote` also gets the quote's absolute path and page with one line: "The key claims to quote this page: confirm it does." Leave the reply format and the verdicts to the agent's own definition; do not restate or narrow them.
3. Read the reply. Split each line on its first three `|` into `N`, verdict, fake-real and reason; strip and lowercase the verdict and fake-real cells. Use only the first line for each `N` that was in the batch you sent. Append each verdict to item N's `verdicts`. `fake-real yes` with `correct` is recorded as `source-conflict`. An item with no usable line or an unknown verdict word gets `ambiguous`; if the Agent call (or fallback pass) fails or errors, every item in that batch gets `ambiguous`. Remember each item's latest fake-real answer. Rerun the gate with the verdicts.
4. If `retry` is empty, go to step 5. Otherwise fix each `retry` item from its reason, checking the source yourself: correct the answer key or the ambiguous question; for fake-real, correct `source`, `page` and `question` to where the question really is per `INDEX.md`, or make it `practice`. Set `dispute` to a few of your own words naming the disputed point (e.g. `step 3`). Keep its `verdicts`. Rerun the gate for the fixed labels, then make one more `mimir:verifier` call with all `retry` items, the same way as step 2, with no second wait line. Read the reply as in step 3.
5. Final gate. First, every item whose latest verifier line said `fake-real yes` becomes `practice` (drop `source`, `page` and `question`, keep `verdicts`): a shown or flagged item never keeps a real label the checker disputed. Then run the gate with all verdicts. Only `show` and `flagged` items reach the student. Any other item is not shown; say once "Dropped K item(s) my checker could not confirm." when K > 0.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |