rem-init ยท diff
git:20260929.3d329f3 to git:20261001.7abf041
19 added, 50 removed. Audit A to A.
---
name: rem-init
- description: The first run of a notebook. Map installed skills, people, organizations and projects with page skeletons, connect sources, rank what matters, and leave an investigation queue. Drives co rem and the mail CLIs.
+ description: Run the first REM notebook investigation so the owner immediately sees useful, cited memory about recent important people, projects and organizations.
---
- # Initialise the notebook
-
- You are building the frame once. Everything programmatic is a command; you
- supply the judgement between the commands. Do not reimplement in prose what a
- command already does. The task supplies the notebook root: include
- `co rem --root "<root>"` in EVERY co rem command below, including follow-ups.
- Read existing pages and `co rem list people --aliases` before creating another identity.
- Do not call `co rem init` recursively or install a background schedule here.
-
- Treat `co rem --help` as the executable workflow guide: it explains task
- selection, observed inputs, expected results, side effects and recovery. Read
- `co rem investigate --help` before investigation and follow actual `Next:`
- commands. This Skill adds evidence judgment; it must not invent a second command
- syntax or turn example page names into assumed records.
-
- Before using the source CLIs, read [CLI.md](CLI.md), beside this file. It gives
- the exact mail, browser and co rem command forms, ID handling, pagination and
- failure recovery. Resolve that path from this Skill's directory, not the task's
- working directory. Run commands through the shell tool; do not invent tool names
- or assume that Gmail, Outlook and `co email` share an inbox or query syntax.
- Use the supplied absolute notebook root in every co rem command; examples below
- abbreviate it for readability. Replace placeholders with observed values.
-
- ## 1. Build the frame without a model
-
- Run `co rem --root '<absolute-notebook-root>' init --days 90`. This command
- creates all four maps and complete page skeletons before returning. It never
- starts a model or investigates a page. It saves mail bodies privately under
- `.state/mail/` and makes per-person references and per-project session indexes;
- the page skeletons never contain raw bodies. If a short metadata-only run is
- needed, pass `--no-mail-archive`. Do not invoke this Skill from that command.
- Connected mailboxes are mapped by default; `--mail` selects explicit mailboxes. Read the init coverage report.
- Use `--skills-dir` for known extra roots (explicit roots replace defaults).
- Existing skills remain at their source paths and existing pages are preserved.
-
- ## 2. Judge the recorded enumeration
+ # Initialize REM
- Read `.state/map.json`, `notes/people-map.md`, `notes/orgs-map.md`, `notes/projects-map.md`, and
- `skills/catalog/index.md`. Counts describe the recorded window, not lifetime
- usage. All correspondents remain unclassified; automated hints never filter
- people by themselves. Classify people, company notices, opportunities and noise
- from evidence. Group repository aliases only when identity is established.
- Write ranking and reasons in `notes/`, preserving source coverage and unknowns.
- Use `scan orgs` and `stub org` for organisation proposals when warranted.
+ Use the notebook root supplied by the task in every command. Read [CLI.md](CLI.md)
+ for source CLI syntax and recovery; resolve it beside this Skill. `co rem --help`
+ and `co rem init --help` are the command contract. Do not recreate the command's
+ mapping or investigation logic in a long prompt.
- ## 3. Stop before investigation
+ Run `co rem --root '<absolute-notebook-root>' init --days 90`. It maps connected
+ sources, saves their private evidence, then uses the configured runner to write
+ the owner's page and investigate recent important people, projects and related
+ organizations. The command announces an estimate before model work. Roughly
+ 20% of a weekly runner allowance is a **target**, not a stopping point: finish
+ the selected investigation even when it takes more, unless the configured
+ weekly safety floor, runner failure, or interruption stops it. Do not pass
+ `--no-investigate` for a normal first run.
- Report mapped pages and source gaps. Suggest an explicit `co rem --root
- '<absolute-notebook-root>' investigate '<record>'` as the next step, owner first
- when identity is verified. Do not execute investigation during init. People,
- projects and skills follow their independent canonical page templates; map and
- investigate use the same headings. No schedule or batch is started here.
+ Read the result and `co rem --root '<absolute-notebook-root>' logs`. Check that
+ the owner's page and selected pages have useful findings and citations, not
+ only empty frames. Follow the returned `Next:` command and retry failed pages
+ with `co rem investigate` or `co rem projects write`. Report source gaps and
+ unfinished pages. Never copy private source bodies into a public document.