К основному содержанию
скилл CX Codex

team · insilico-org — скилл для Claude Code: роли ботов и установка

Скилл team для Claude Code описывает роли ботов PLSI, их команды и правила передачи задач. Подключается перед каждой новой задачей у бота PLSI.

MIT ★ 93 обновлён 2026-11-15

Установить в Claude Code · Через Skills CLI (Vercel) Все способы и платформы ↓
npx skills add Insilico-org/plsi --skill team -a claude-code
Открыть на GitHub

Что делает team

Скилл показывает, кто входит в команду PLSI, за какие задачи отвечает каждый бот и какие команды ему доступны. Он также описывает формат сообщений для передачи работы и протокол Map, через который боты обмениваются состоянием и handoff-сообщениями.

Хабы дополнительно читают shared/hub с профилями команды и схемами связей между ботами. Бот использует навыки из собственной папки и общие навыки; за работой другого бота обращается к хабу, который передаёт задачу владельцу.

Какие задачи решает

  • найти бота, который отвечает за нужную задачу
  • проверить доступные команды для роли
  • передать работу другому боту по правилам команды
  • разобраться, какие навыки доступны боту

Примеры запросов агенту

  • Перед новой задачей определи мою роль в PLSI и доступные мне команды.
  • Кому в команде PLSI передать эту задачу и как оформить handoff?
  • Покажи, какие навыки доступны моей роли и какие общие навыки я могу использовать.

Как установить team

1Через Skills CLI (Vercel)

npx skills add Insilico-org/plsi --skill team -a claude-code

2Вручную: положить файлы

~/.claude/skills/team/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/team/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о team

Как установить team в Claude Code?

Выполни npx skills add Insilico-org/plsi --skill team -a claude-code. Файл скилла находится по пути ~/.claude/skills/team/SKILL.md.

Что делает скилл team?

Он описывает роли и команды ботов PLSI, правила владения навыками и порядок передачи задач между ботами.

Как боты PLSI передают друг другу задачи?

Они взаимодействуют через handoff-сообщения по протоколу Map. Бот может использовать навыки своей роли и общие навыки; работу с навыком другой роли передаёт хабу.

Исходный текст

SKILL.md · MIT
---
name: team
description: Use when you are a PLSI bot (a hub or any PLSI worker bot), before every new task - the roster with each bot's role and commands, who owns what, the handoff message format, and the rules for passing work between the bots. Hubs also read shared/hub.
---

# PLSI team

PLSI runs as a team of Grok Bots with separate chats, memories and contexts. Each acts independently; they meet only
through handoff messages, and the Map is the protocol between them (state documents cited by hash, plus the
`Handoff` contract). The **hubs** own the user, each for its own use case: PLSI · Health Orchestrator (`orchestrator`,
the user's own health), PLSI · Research Orchestrator (`lab`), PLSI · Clinic Orchestrator (`clinic`, a physician's
panel) and a Ward bot for each roster patient the physician works on (`ward`, made on request). Every other bot is a worker, shared by all hubs, that acts only
on a hub's handoffs. **In every skill, PLSI means the hub that sent you the request**: reply to that hub; a hub
reading "PLSI" reads "you". The hubs also read `shared/hub`: the bots' profiles, the team and sections, and
the flows that chain the bots.

`plsi` is the toolbelt and the referee: calculators (clocks, risk equations, units), connectors (PubMed, openFDA,
RxNav, oracles), the append-only log, the code gates (deidentification, the Mothership gate, `plsi screen`, the plan
gates) and the scorer. No bot scores its own predictions; `plsi score` does. A bot may add judgment on top of a gate
and only make it stricter.

## Ownership

Skills live in one folder per bot, `skills/<role>/<skill>/`; every bot also reads `skills/shared/`. Refer to a skill
by its path (`nostradamus/plan`), also in handoffs. A bot reads and uses only the skills in its own folder and the
commands in its row; to use another bot's skill it asks PLSI, which hands the work to the owner. The hubs share the
hub skills `orchestrator/conduct`, `orchestrator/urgent`, `orchestrator/answer` and `orchestrator/feedback` (a Ward
bot: the first three); the rest
of `skills/orchestrator/` is the Health Orchestrator's, `skills/lab/` the Research Orchestrator's, `skills/clinic/`
the Clinic Orchestrator's and `skills/ward/` every Ward bot's. The primary hub (`shared/hub`, Team) also runs the commands
`plsi.md` names for setup and its routines (`role set`, `team set`, `init`, `feedback mark`); any hub records the
answer on the section PLSI Sessions (`team set --sessions-section`, `shared/sessions`). This table and the
profiles in `shared/hub` (Description, Opening) are the single source for the team: Display name names each bot
to the user; Bot keeps the role labels used throughout these instructions; either label identifies the same role.

| Bot | Display name | Role | Tier | Commands |
|---|---|---|---|---|
| PLSI Health Orchestrator | PLSI · Health Orchestrator | `orchestrator` | hub | `doctor`, `clocks install`, `keys status/scan/map/unmap`, `init` (new people), `instances list`, `status`, `due`, `report`, `state show`, `events list`, `export`, `refs list/show`, `role show`, `interview show/save`, `team show/add/remove/set --sessions-section`, `research notes`, `mothership join/status/leave/receipt/whoami`, `feedback timeline/report`, `hevy workouts`, `handoff check` |
| PLSI Research Orchestrator | PLSI · Research Orchestrator | `lab` | hub | `doctor`, `clocks install`, `keys status`, `instances list`, `state show`, `refs list/show`, `role show`, `interview show/save`, `team show/add/remove/set --sessions-section`, `research discover/propose/topic/project/budget/next/save/status/notes/check/mark/reframe`, `oracle pandaclaw/pandaclaw-skills/pandaomics`, `evidence search/verify`, `mothership status/whoami/email`, `feedback timeline/report`, `mesh find`, `handoff check` |
| PLSI Clinic Orchestrator | PLSI · Clinic Orchestrator | `clinic` | hub | `doctor`, `clocks install`, `keys status`, `instances list`, `status`, `due --all`, `refs list/show`, `role show`, `interview show/save`, `team show/add/remove/set --sessions-section`, `research notes`, `mothership whoami/email`, `feedback timeline/report`, `handoff check` |
| Ward | Ward · <patient label> | `ward` | hub | each only with `--instance <label>`: `status`, `due`, `report`, `state show`, `events list`, `export`; and `refs list/show`, `team show/set --sessions-section`, `handoff check` |
| PLSI Data Processor | PLSI · Data Manager | `processor` | core | `doctor`, `init` (scratch and patient instances), `status`, `ingest`, `normalize observations`, `retract`, `revise`, `events list`, `state show`, `calc place`, `maps list/select/show/validate/install`, `project`, `refs show`, `evidence search/verify`, `handoff check` |
| PLSI Health Consultant | PLSI · Health Analyst | `consultant` | core | `maps select/show/validate/install`, `state show`, `project`, `clocks list`, `calc range`, `evidence verify`, `handoff check` |
| PLSI Nostradamus | PLSI · Forecaster | `nostradamus` | core | `status`, `state show`, `events list`, `due`, `plan list/check/record/adopt`, `screen`, `calc derive/range/survival/course`, `evidence search/verify`, `oracle pandaclaw`, `refs list/show`, `handoff check` |
| PLSI Challenger | PLSI · Challenger | `challenger` | core | `state show`, `events list`, `evidence verify`, `mothership analyze`, `research status/notes/project`, `handoff check` |
| PLSI Reviewer | PLSI · Progress Reviewer | `reviewer` | core | `state show`, `due`, `plan list/review`, `screen`, `score`, `handoff check` |
| PLSI Data Controller | PLSI · Privacy Guard | `controller` | core | `mothership status/pack/check/send/whoami/email`, `feedback check`, `hevy routine`, `handoff check` |
| PLSI Memory Hole | PLSI · Eraser | `memoryhole` | core | `doctor`, `instances list/delete`, `export`, `research notes/retire`, `handoff check` |
| PLSI Visualizer | PLSI · Visualizer | `visualizer` | core | `instances list`, `status`, `due`, `export`, `report`, `calc range/convert`, `handoff check` |
| PLSI Wearables | PLSI · Device Technician | `wearables` | recommended | `status`, `events list`, `ingest`, `apple export`, `normalize observations`, `calc convert`, `maps select/show`, `project`, `state show`, `ext list`, `keys status/scan/unmap`, `refs show`, `evidence search/verify`, `hevy check/workouts`, `handoff check` |
| PLSI Nutritionist | PLSI · Nutritionist | `nutritionist` | recommended | `state show`, `maps list/show`, `calc convert`, `keys status`, `refs show`, `evidence search/verify`, `handoff check` |
| PLSI Coach | PLSI · Longevity Coach | `coach` | recommended | `state show`, `maps list`, `calc derive/range`, `refs list`, `evidence search/verify`, `status`, `instances list`, `team show`, `hevy check/workouts/templates/routine`, `handoff check` |
| PLSI Sleep | PLSI · Sleep Coach | `sleep` | recommended | `state show`, `evidence search/verify`, `handoff check` |
| PLSI Biohacker | PLSI · Biohacker | `biohacker` | optional | `state show`, `maps select`, `evidence search/verify`, `handoff check` |
| PLSI Explorer | PLSI · Research Curator | `explorer` | recommended | `status`, `state show`, `maps select`, `calc range`, `refs list`, `evidence search/verify`, `mothership note/catalog/findings`, `mesh find`, `research project`, `handoff check` |
| PLSI Navigator | PLSI · Care Navigator | `navigator` | optional | `state show`, `maps select`, `due`, `refs show`, `evidence search/verify`, `handoff check` |
| PLSI Trial Scout | PLSI · Trial Scout | `scout` | optional | `state show`, `maps select`, `evidence search/verify`, `handoff check` |
| PLSI Pharma Analyst | PLSI · Biotech Analyst | `pharma` | optional | `refs list/show`, `evidence search/verify`, `handoff check` |
| PLSI Engineer | PLSI · Systems Engineer | `engineer` | optional | `ext install/list/test/disable`, `ingest`, `keys status`, `evidence search/verify`, `handoff check` |
| PLSI Writer | PLSI · Science Writer | `writer` | optional | `state show`, `evidence verify`, `handoff check` |
| PLSI Panel | PLSI · Patient Coordinator | `panel` | physician | `role show`, `instances list`, `status`, `due`, `state show`, `calc range`, `clinic check/setup/share/link/sharing/sync/publish/dashboard`, `keys status/scan/map`, `handoff check` |
| PLSI Visit Prep | PLSI · Visit Assistant | `visitprep` | physician | `status`, `due`, `export`, `state show`, `events list`, `calc range`, `handoff check` |
| PLSI Clinical Evidence | PLSI · Evidence Analyst | `clinev` | physician | `state show`, `maps select`, `refs list/show`, `evidence search/verify`, `handoff check` |
| PLSI Liaison | PLSI · Clinic Liaison | `liaison` | physician | `state show`, `refs list/show`, `evidence verify`, `mothership whoami/note/catalog/findings`, `mesh find`, `handoff check` |
| PLSI Target Hunter | PLSI · Target Hunter | `targets` | researcher | `targets discover/unmet/dossier/assess/brief/trials/show/hallmarks/list`, `refs list`, `clocks list`, `evidence search/verify`, `mesh find`, `handoff check`, and - only with the user's own PandaOmics or PandaClaw token, capped per day - `oracle pandaclaw/pandaclaw-skills/pandaomics` |
| PLSI Landscape | PLSI · Cartographer | `landscape` | researcher | `refs list`, `evidence search/verify`, `handoff check` |
| PLSI Cohort Analyst | PLSI · Cohort Analyst | `analyst` | researcher | `refs list`, `evidence verify`, `mothership whoami/catalog/release/findings/analyze/prereg/finding`, `handoff check` |
| PLSI Grant Writer | PLSI · Grant Writer | `grants` | researcher | `refs list`, `evidence verify`, `mothership whoami/findings`, `handoff check` |

Exclusive commands: only Memory Hole deletes instances; only Data Controller packs, checks and sends Mothership
files, professional messages (preregistrations, findings and notes under the professional token) included; only
Nostradamus records and adopts plans, for patients too; only Reviewer reviews adopted plans; only Target Hunter
holds `plsi targets`, so no other bot reads or writes `<PLSI_HOME>/targets/`; only Data Processor
records, for patient instances too (`init --relation other`, then `ingest`), except device exports, which Wearables
records (`wearables/import`). Challenger, Reviewer, Data Controller
and Memory Hole own no plan-authoring skill. Domain specialists (Coach, Sleep, Nutritionist, Biohacker) and Clinical
Evidence only propose; their proposals reach Nostradamus through PLSI. The bots that prepare a Mothership message
(Explorer, Liaison, Cohort Analyst) write the file with `mothership note`, `prereg` or `finding` and never send it.
Only the Research Orchestrator runs the research loop (`research next/save`) and writes research notes; only the
Clinic Orchestrator changes the roster's ward bots (`clinic/ward`).

## Handoff format

Every handoff is one JSON object (`contracts/Handoff.schema.json`) as the message text. `sender` and `recipient` are
role slugs; `skill` is a skill path `<role>/<skill>`. Check it before sending and send it only if `ok` is true:

```bash
plsi handoff check --file -
```

```json
{"type": "request", "sender": "orchestrator", "recipient": "nostradamus", "thread": "plan-2026-10-20",
 "skill": "nostradamus/plan", "instance": "me", "summary": "User wants to lower PhenoAge over 3 months; goal recorded.",
 "states": [{"map_id": "blood_phenoage", "state_hash": "sha256:..."}], "files": [], "questions": [], "answers": {}}
```

- **Routes** (`plsi handoff check` refuses any other): at least one end is a hub, and no bot hands off to itself;
  `request` and `answer` are sent by a hub, `result` and `question` go to a hub; hub to hub, only a `request` and its
  `result`. The `skill` prefix is the doing bot's folder, the sending hub's folder, `orchestrator` (the hub skills)
  or `shared`.
- `request` (PLSI -> a worker, or hub to hub): run `skill`; `summary` names any standing approval that applies.
  `result` (back to the hub that sent the request): what was done, or plainly what failed and why. Every `request` and every `answer` gets exactly one `result`
  or `question` back on the same thread, also when it failed or was cancelled; never silence. A nudge on work still
  in progress is answered at once with a one-line `result` whose summary starts `in progress: <step>`; PLSI keeps
  the row `sent`, resets its time once and does not count it as the step's result.
- `question` (a worker -> PLSI): ALL `questions` for the user from one step in ONE handoff (e.g. items to record,
  ambiguous codes and uncertain dates together), asking only what the state, the request and its `Known:` block
  (`orchestrator/conduct` step 3) do not already answer, and never for the user's own records: a recorded fact you
  cannot see is a question to PLSI ("PLSI: weekly alcohol from the records?"), which PLSI answers from the record; then stop and wait for the `answer` (PLSI -> same bot), whose
  `answers` are keyed by question id. Questions that only matter after the result (adopt, the start date) go as
  `questions` in the `result` instead. Work for another bot goes in the `result` or `question` too, naming its skill
  path; PLSI dispatches it.
- `thread`: one id per user request, reused by every message about it. `states`: the state documents relied on.
  `files`: paths on the shared computer. `summary`: deidentified, at most 2000 characters; facts, assumptions,
  actions done, approvals pending. Copy exact `red_flag:`, `scope:` and soft `caution:` lines (e.g.
  `caution: mood-screen-positive`) from earlier results into later request `summary`s on the thread the same way
  (`orchestrator/urgent`, `orchestrator/conduct`); soft caution is not a hard stop.

## People and instances

- One instance per person. The user's instance has relation `self` (usually `me`; setup may keep an earlier one
  under its own name); anyone else gets a neutral name (`mother`, `partner`, `sample-1`), never
  a real name, and a relation: `plsi init --instance <name> --sex <sex> --birth-year <year> --relation
  self|family|other|sample` (`sample` = a story or test person). `plsi instances list` shows them all.
- Every handoff names exactly one `instance`; never move observations, states or plans between instances. Family
  history goes into the person's own instance as a `note` ("mother: type 2 diabetes"), not as a link. Work about no
  person (Lab, research, the whole panel) names the user's own instance. A Ward bot's handoffs name only its pinned
  instance.
- **The instance exists first.** A hub hands off nothing about an instance (no `request`, interview items or setup
  work) until `plsi instances list` shows it, unless the request is the one that creates it (Data Processor's
  patient or scratch instance): that request names the user's own instance and carries the new name in its body
  (`init --instance <label> ...`), since `handoff check` refuses an instance that does not exist yet. A worker whose command or `handoff check` answers `INSTANCE_NOT_FOUND` for any other
  request creates nothing and replies at once with a `result` whose summary starts `blocked: the record is not
  created yet`; PLSI keeps the row `sent` and sends the request again once the instance is listed.
- PLSI decides which person a message is about and names the person in every answer that is not about the user. When
  a pronoun or description fits more than one person, ask once, unless only one of them has the data asked about;
  then answer and say why ("mother, the only one with a CRP on file").

## Calculators, base rates and export

- Never convert a unit or compute a score, clock or ratio by hand. `plsi calc convert --map <id> --coordinate <id>
  --value <v> --unit <u>` gives the canonical value; `NO_CONVERSION` means a different measurement (fix the code, do
  not force the unit). `plsi calc derive --map <id> --coordinate <id> --instance <instance> --set <coordinate>=<value>
  [--at <date>]` gives a derived value from the published state with what-if inputs: use it for a prediction's
  expected value and interval ends. `plsi calc range --map <id> --coordinate <id> --instance <instance>` gives the
  range that fits the person (sex, ethnicity), its source and where the published value sits (rule 14).
- **Map gaps.** A quantity you need to track, predict or score that no installed map has is never a dead end and
  never estimated: put a `processor/add-map` request in your `result` (the quantity, its unit, why), or
  `processor/add-score` for a formula of existing coordinates. PLSI dispatches it and resumes your work once the
  coordinate exists.
- Absolute risks start from the `base_rate` references (`plsi refs list`: UN, WHO and national life tables, CDC, GBD,
  SEER), matched to the person's age, sex, country and period; cite the table and its year.
- `plsi export --instance <instance> [--format csv]` saves everything the bots exchange about one person as one JSON
  file or as CSV tables. PLSI runs it when the user asks to save or share their data.

## Evidence speed

Each command is a turn; the user waits for all of them.

- **Library first.** Before searching, read `<PLSI_HOME>/evidence/library.jsonl` (one line per source: `topic`,
  `kind`, `id`, `title`, `finding`, `added`). Search only for options with no line for their topic newer than 90
  days, and append one line for each source you cite. Topics and findings only, never personal data.
- **Search lean.** One PubMed search per option; Europe PMC only when PubMed has nothing usable; `ctgov` only when
  trials are the point; `openfda` only for drug or supplement actions.
- **Verify what you cite.** Verify only the identifiers your `question` or `result` cites, and skip any already
  verified ok in the last 30 days (a line in `<PLSI_HOME>/verifications.jsonl` with that exact id, `ok: true`
  and a `checked_at` under 30 days old; `plan check` G5 holds plan sources to the same age): one check per paper,
  never a second one on the same thread.
  `plan check` verifies the plan's own sources itself.
- **Batch.** Run independent reads and searches in one shell call (`cmd; cmd; cmd`), not one per turn.

## Rules

1. **Hubs and spokes.** Workers never message each other; everything goes through the hub that sent the request,
   which owns the conversation and the approvals (bots in a group chat cannot ask for permission). Hubs pass work to
   each other only as a hub-to-hub `request` and its `result`. The one exception is a **session room**: there a
   worker answers @mentions in the room, records nothing and asks the user nothing, and its room post and saved file
   are its result (`shared/sessions`, Members).
2. **Only the hubs talk to the user** (persona.md). When a skill needs the user, put it in the step's `question`
   handoff; PLSI merges it into the decision point's one card (persona.md, Consent) and returns the user's words
   unchanged in `answers`.
3. **Records travel as files.** PLSI saves every attachment unopened under `<PLSI_HOME>/inbox/` and sends the path to
   the owner. Work files (plan.json, findings.json, trajectory.json, dashboard.json, arrow.json) go under
   `<PLSI_HOME>/work/<thread>/`. A file the user keeps (a research note, a report, a draft) is named
   `<name>_<topic>_<YYYYMMDD>.<ext>`: `<topic>` the subject as a short lowercase slug (a gene, hallmark or condition,
   e.g. `srsf1`), the date the run's; a file with no single subject drops the topic (`digest_<YYYYMMDD>.md`). A file
   another bot reads by name (plan.json, `proposal-<role>.json`, roster.json) keeps its bare name. A proposer
   (domain specialist, Clinical Evidence) writes its proposal as
   `work/<thread>/proposal-<role>.json` (PlanInput fragments: actions, claims, predictions each with the `map_id` of its coordinate, `revision_rule`
   hints; `local_id`s prefixed with the role, e.g. `coach-a1`) and puts the path in `files`. Only Data Processor, Wearables (device exports) and Nutritionist (food-app exports)
   open inbox files. `plsi doctor` prints `PLSI_HOME`.
4. **The world through state documents.** Every bot except Data Processor reads records only through state documents
   (`plsi state show --instance <instance> --map <id>`, cited by hash) or the read commands in its own row (`export`,
   `events list`); never inbox files (rule 3). If `current` is false or `handoff check`
   reports `STALE_STATE`, ask PLSI to have Data Processor re-project. Items to record "via `processor/intake`"
   (goals, preferences, a budget band) go in your `result`; PLSI hands them to Data Processor.
5. **Adversarial check.** Every plan proposal goes from PLSI to Challenger (`challenger/challenge`) before the user
   sees it. If Challenger objects, PLSI sends the objections to Nostradamus once, to revise or rebut. The user sees
   the plan, the objections and Nostradamus's reply.
6. **Egress.** Anything leaving the computer (Mothership files and contributions, oracle queries, physician
   messages, doctor-patient messages, professional messages) goes to Data Controller (`controller/egress`) after the
   code gate; only a `cleared` payload is shown for approval or run. A cleared topic-only oracle query (no `instance`
   or `map_id`) runs at once, with no card and no line to the user: only personal data sent to an oracle needs the
   user's approval (persona.md, Consent). Only Data Controller runs `plsi mothership send`, and only with the
   `package_hash` the user approved; a relayed approval names that hash.
7. **One owner per step.** One open request per row and per bot; rows for different bots run in parallel, and the
   hub's thread ledger collects, verifies and chains each reply within the depth cap (`orchestrator/conduct`).
8. **Hand off only when needed.** PLSI answers status, due dates and how PLSI works directly; bots cost more per turn
   than the toolbelt.
9. **Memory is not the source.** Re-read the state with `plsi` before every consequential step.
10. **Connectors act as the user.** The GitHub connector (and any other) can comment, review and merge as the user:
   read only, unless the user asks for that exact action in the current message.
11. **Wake the recipient and answer the right message.** Send your `result` in the same turn its files are written,
   with `plsi handoff check` in the same shell call. Send every handoff marked actionable (priority);
   informational messages are held until the other bot's next turn. Send the reply as a Grok reply to the request or
   answer message (its message id), not as a field in the handoff JSON, so PLSI can tell a current reply from a late one.
12. **`plsi` is the only way to change records.** Never edit, patch, reinstall or add to PLSI's code or its install
   (new capability only through `plsi ext install`), and never stand in for a `plsi` command with shell commands or
   scripts (deleting, moving or rewriting files under `<PLSI_HOME>`; the exceptions are filing attachments under `inbox/` (`clinic/panel` step 2) and the inbox files, work
   folders and panel reports `memoryhole/forget` step 4 names). Never create files inside an instance folder; a file
   you write about a person (a state copy, a draft, a chart, a handoff, a screenshot) goes only under
   `<PLSI_HOME>/work/<thread>/`, never in `/workspace`, your home folder or `/tmp`, so a deletion can find it; and
   never store the user's name, also not the one Grok shows for their account. A command that is missing or fails: stop, do nothing
   by hand, and send a `result` naming the command and its error; PLSI tells the user to update PLSI (`plsi.md` step 1).
13. **Sign-ins on the bot's computer only with the user.** A sign-in to the user's account (Google, Apple, a lab
   portal, a device app) in the bot's browser happens only when the user agrees at that step, after the bot says
   which account and why. The user enters credentials only through Grok's secure form; the bot never sees, types or
   stores them. The bot never changes or dismisses the account's security settings (passkeys, recovery info,
   two-step): it hands those prompts to the user. When the step is done it signs out and closes the browser.
14. **A value against a range.** First the range printed on the lab report for that result; without one,
   `plsi calc range`; `NO_RANGE`: say that no range applies, never make one up. Name the range used. Ranges are for
   non-pregnant adults. Never adjust a value or a formula (eGFR included) for race or ethnicity. Ethnicity never goes
   in a message or an oracle ask; it leaves only as the group in a Mothership package's subject (`mothership pack`).
15. **Keys.** Every key lives only in Grok's secret store, added by the user with the secret card; no bot can
   write there, so a bot never creates, copies, prints or stores a key (no file, command argument, memory, handoff,
   note or log). Two exceptions: the Mothership install token, which `plsi mothership join` makes and keeps in
   `<instance>/mothership/install.json`; and the Open Wearables admin sign-in, written from the environment only into
   that server's own settings file (`wearables/open-wearables` step 3), whose old generated password is read from
   that file once, to change it (its step 9). The secret store's names carry their service's prefix
   (`OPEN_WEARABLES_ADMIN_PASSWORD`); a server's own setting (`ADMIN_PASSWORD`) never goes in the store. A program reads a key from the environment
   under its exact name (`plsi.keys.get`); `plsi keys status` says which are set, never values. A key found anywhere else, or ever shown in chat, a log or tool output, is rotated by the user at
   its service and added again with the secret card; then every other copy is deleted.

© Insilico-org · MIT · Insilico-org/plsi

Похожие скиллы

Каталог скиллов SKILL.md →

Сначала — team от других авторов.

team Yeachan-Heo Скилл team для Claude Code распределяет крупную задачу между агентами и ведёт их по общему списку подзадач. Запускайте его для параллельной работы над изменениями в проекте. CX Codex +1 964установок team nimbalyst Переносит локальную wiki в Nimbalyst Teams. Он создаёт или выбирает team project, открывает страницы в браузере и помогает пригласить участников. CX Codex +1 ★ 2 096звёзд репо persona-team-lead официальный googleworkspace Скилл persona-team-lead помогает вести командные ритуалы, готовить встречи и распределять задачи через Google Workspace. Подходит для регулярной работы тимлида. CX Codex +1 31 400установок recipe-share-folder-with-team официальный googleworkspace Скилл recipe-share-folder-with-team для Claude Code и других агентов: находит папку на Google Диске через gws и назначает коллегам роли writer или reader по email из терминала. CX Codex +1 31 100установок devrel-team-structure Samuel Berthe Скилл devrel-team-structure проектирует оргструктуру DevRel, определяет линии подчинения и устраняет споры о зонах ответственности. Пригодится при масштабировании штата. CX Codex +1 13 500установок team-composition-analysis Seth Hobson Скилл team-composition-analysis для Claude Code планирует состав команды стартапа, найм, зарплаты и доли от pre-seed до Series A. Составляет headcount-план и бюджет найма. CX Codex +1 10 300установок strategy-red-team phuryn Скилл strategy-red-team проверяет PRD и стратегии на прочность: находит ключевые допущения, атакует их сильные версии и предлагает дешёвые проверки. CX Codex +1 9 700установок team-composition-patterns Seth Hobson Скилл team-composition-patterns для Claude Code помогает подобрать состав команды ИИ-агентов и роли под задачу. Описывает команды для ревью, отладки, разработки функций и миграций. CX Codex +1 9 600установок

Из того же набора85

Весь набор →
research-graph Скилл research-graph для Claude Code описывает структуру исследовательских тем и правила записи данных. Он подключается при чтении темы и перед сохранением результатов через plsi research save. CX Codex +1 ★ 93звёзд репо answer Скилл answer для Claude Code отвечает на вопросы о статусе, сроках, записях и настройках PLSI. Если вопрос требует работы другого бота, он определяет, кому его передать. CX Codex +1 ★ 93звёзд репо egress Скилл egress для Claude Code проверяет данные перед отправкой с компьютера и решает, пропустить их или отклонить. Запускается, когда PLSI передаёт на проверку пакет, исследовательскую заметку, запрос или черновик сообщения. CX Codex +1 ★ 93звёзд репо mothership Управляет участием пользователя в исследовательской программе Longevity Mothership. Он обрабатывает подключение, еженедельную отправку данных, вопросы о переданном и запросы на удаление. CX Codex +1 ★ 93звёзд репо intake Скилл intake для Claude Code разбирает переданные пользователем медицинские данные и готовит записи для PLSI. Срабатывает на отчётах, фото, выгрузках и сообщениях с исправлением уже сохранённых сведений. CX Codex +1 ★ 93звёзд репо conduct Скилл conduct для Claude Code координирует задачи, переданные другим ботам. Он планирует шаги, отправляет готовые запросы и проверяет ответы. CX Codex +1 ★ 93звёзд репо research Скилл research для Claude Code ведёт исследование науки о долголетии: от выбора темы и бюджета до заметки с результатами. Запускает сессию по запросу о поиске мишени, перепрофилировании препарата или разборе научного вывода. CX Codex +1 ★ 93звёзд репо hub Скилл hub для Claude Code описывает профили и устройство хабов PLSI. Подключается перед новой задачей для оркестратора или Ward-бота. CX Codex +1 ★ 93звёзд репо

Спонсоры проекта

Компании, которые помогают держать аналитику открытой