forma
/forma
Instaluje lub aktualizuje w projekcie protokół Forma z pięcioma węzłami i sprawdza każdy składnik osobno, nie nadpisując istniejącego.
- Autor
- FormaMindPro ↗
- Repozytorium
- formamind ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 11.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 719
- Skrypty
- 261
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 11.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 261 skryptów w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
- Paczka skrócona: część plików jest zbyt duża albo wykonywalna. Pełny folder jest w repozytorium.
Skrypty: adapters/claude/dashboard/floor.cjs, adapters/claude/dashboard/subagent-transcript.cjs, adapters/claude/dashboard/subagent-transcript.test.cjs, adapters/claude/dashboard/subagents-log.cjs, adapters/claude/forma-adapter.cjs, adapters/claude/hooks/auto-graphs.sh, adapters/claude/hooks/check-card.sh, adapters/claude/hooks/check-clean-zone.sh, adapters/claude/hooks/check-credentials.sh, adapters/claud
Co robi i kiedy
Instaluje lub aktualizuje w projekcie protokół Forma z pięcioma węzłami i sprawdza każdy składnik osobno, nie nadpisując istniejącego.
Wyzwalacze z opisu
Oryginalny opis (EN)
Installs or updates the "Forma" protocol - the five nodes (Intent/Spec/Kit/Run/Core) - into the current project. Invoke when the user asks to deploy/install/connect/update the "Forma" protocol in a project ("forma update", "update forma" and the like), or explicitly invokes /forma. A repeated call on an already installed project is safe - it checks each goal separately (SKILL.md, step 1), never silently overwrites what already stands. The interviews (terminal + browser) and the dashboard with the graph are part of the core and are installed by the same call, without a second step. Two detachab
Jak działa
- Installation - one call of the installer
- Skeleton translation (only for unbundled languages, first installation only)
- What the installer does (reference)
- What this skill does not do
Podgląd SKILL.md
---
name: forma
description: Installs or updates the "Forma" protocol - the five nodes (Intent/Spec/Kit/Run/Core) - into the current project. Invoke when the user asks to deploy/install/connect/update the "Forma" protocol in a project ("forma update", "update forma" and the like), or explicitly invokes /forma. A repeated call on an already installed project is safe - it checks each goal separately (SKILL.md, step 1), never silently overwrites what already stands. The interviews (terminal + browser) and the dashboard with the graph are part of the core and are installed by the same call, without a second step. Two detachables remain: the project template - by the human's choice; the kanban board - an external editor extension (step 9).
---
# forma - installing the "Forma" protocol
Copies the protocol into the current project in two layers. **The core** (`core/`) is one for all engines and is always installed: the law `AGENTS.md` §1-7, `manual/`, the `project/` skeleton, the dashboard with the economy module (`dashboard/economy.cjs`, `tally.cjs`, `spend-line.cjs`), the board (`board/`, the blank `.devtool/features/`), both interviews (`skills/`), `living/`. **An engine adapter** (`adapters/<engine>/`) - §8, roles, hooks, scripts, skills and its own economy adapter; the chosen one is installed. One is ready - `adapters/claude/` (Claude Code, default); `adapters/codex/` and `adapters/gemini/` are carried over as they are and marked `BETA.md` - **installed as beta**: they have no conformance test yet. In the composition: the common rules, the five agent roles, the delete-guard hook, the schema's universal documentation (`manual/`) and an empty skeleton of the project structure (`project/`, common to both environments). It invents nothing and adapts nothing to the project - carries as is; filling it for a specific matter (thresholds, access, epics, goals) happens later, at the opening of the first goal, by `Intent` itself in conversation with the human (`agents/intent.md`, "Housekeeping").
## Installation - one call of the installer
Do not copy files by hand - that is the job of this same plugin's installer `bin/forma.cjs` (the same one as `npx github:FormaMindPro/formamind init` from a terminal). The order:
1. **Ask the human four questions in the chat, one at a time** (on a repeated installation - the same ones; the default answers fit):
- **engines** - Claude Code (ready); Gemini - "beta" (`--engines claude,gemini`); Codex **sleeps** - the registry `core/engines.json` (`dormant`), not offered in the menu, `--engines codex` refuses, step 3c and the Codex reconciliation are skipped. Enabling and disabling - the `engine-switch` skill (`node .claude/scripts/engine-switch.cjs wake|sleep <engine>|status`): removes the marker, rebuilds the engine's data, runs the reconciliation. Default `claude`;
- **the project template** - `none` (the own route is developed in the interview, default) or a name from `templates/` (step 8 below - a fork, decide together with the human);
- **the kanban board** - `auto` (find the CLI `code`/`antigravity-ide`/`antigravity`, install Kanban Markdown where absent) or `skip`;
- **the language of the project documents** - `en` (default, the skeleton stays English as is) or the code of any other language (`ru`, `fr`, `de`…): the agent translates the skeleton after installation, the "Skeleton translation" step. Hook messages go in the project language if a catalog `core/i18n/messages/<code>.json` exists (currently `en`, `ru`); for the others - in English. First installation only: an existing `project/` does not change its language.
2. **Choose the installer's source - it is not the plugin cache by default.** If the project has a clone of the protocol `.forma/protocol` (`.forma/protocol/bin/forma.cjs`), update from it: first `git -C .forma/protocol fetch --tags` and `merge --ff-only origin/main`, the command - `node .forma/protocol/bin/forma.cjs init …`. This skill's directory (the plugin cache) may be older than what is installed: version numbering was reset once, and by semver an old cache looks newer. The installer itself verifies the protocol line (`line` in `package.json` and the manifest) and the version, refuses with code 3 on a downgrade and names the command from the clone; `--allow-downgrade` - only on the human's direct request. No clone - install from the skill's directory.
3. **Run at the project root:** `node <path-to-installer>/bin/forma.cjs init --engines <engines> --template <template|none> --board <auto|skip> --lang <code> [--adopt|--no-adopt]`. All four flags given - the installer asks nothing. **First installation only, when the root already has foreign folders and files** (not the protocol root): ask the human whether to move them into `project/source/` (step 0b), and pass the answer with the flag `--adopt` or `--no-adopt`; without the flag in a terminal the installer asks itself, without a terminal - it will not touch them and will name what it found in the report.
4. **Show the human the installer's report as is:** what was synchronized, whether `project/` was touched, the Kanban Markdown lines (`✓ ok` / `✓ installed` / the command for a manual install), the installation integrity line (`verify-install`). The report not `✓` - take it apart and report (step 8b).
## Skeleton translation (only for unbundled languages, first installation only)
The installer checks if a bundled scaffold package exists in `templates/_scaffold/<code>` (currently `en` and `ru`). For bundled languages, the pre-translated skeleton is laid directly upon installation and no translation is needed. For unbundled languages (such as `fr`, `de`), the installer lays the English skeleton, records the language name in `project/config/PROJECT.md`, and creates `.forma/translation-pending.json` (`from`, `to`, a closed list `files`). The translation is the agent's work, once:
1. No `.forma/translation-pending.json` - no translation needed, no step.
2. Open a card (`node .forma/board/new-card.cjs`), kind `tooling`, epic `protocol`; the executor - `run-text`; in the card - the `files` list from the file, as a closed enumeration.
3. Translate **text only**: do not touch the machine keys - the card zones' headings and fields, file and directory names, codes (`route-N`, `goal-*`, `card-NNN`), statuses from closed lists, JSON keys, `` `paths/in/backticks` `` and `[[name]]` links, the nodes' names (`Intent`, `Spec`, `Kit`, `Run`, `Core`). The invisible anchors `<!-- k:… -->` in `PROJECT.md` and the zones' headings are keys too: keep them as is, parsers read the fields by them, not by the caption text.
4. Compare the skeleton with the English original from `core/project/`: the same files, the same headings in number and order, the same keys/codes, the same tables in rows and columns. A divergence - by address and expected value (prohibition 13), an edit, a repeat.
5. The comparison green - delete `.forma/translation-pending.json`, write a line into the card's history, then - `project/config/SETUP.md`, step 1.
Hook messages are needed in that language - add `core/i18n/messages/<code>.json` with the same codes as in `en.json` (a test compares the sets); without it the hooks print in English, nothing breaks.
The mark `.forma/translation-pending.json` still in place - the skeleton is not fully translated; do not start the interview until it is removed.
A repeated run - an update: the schema's configuration is overwritten, `project/`, the board `.devtool/features/`, `living/` and the recorded template's files are not touched. The steps below are a reference of what the installer does; walk them by hand only for the adapters that are not ready, on the human's direct request (3b, 3c).
## What the installer does (reference)
0. **First create or preserve the target project's Git boundary.** Do this step at the target project's root, before copying any file from the template. Run `node <skill-dir>/../../scripts/initialize-project-git.cjs .`. It is a Node script with no dependency on PowerShell or `sh`; Node is already required by the protocol for its scripts and dashboard and is available in all supported engines.
The script first checks that `git` is available in `PATH`. If Git is unavailable, **stop at once and tell the human** that Git is needed for the installation; copy no protocol file and create no `project/`. If the root has no `.git`, the script runs `git init` exactly in that root and creates a basic `.gitignore` only when no such file exists yet. If a `.git` already exists (including a worktree file), the script touches neither history nor remote. In both cases the script keeps environment secrets per prohibition 15: appends the `.env` line into `.gitignore` if absent (changes nothing else); creates `.env` at the root from `core/.env.example` only if `.env` does not exist yet - an existing one with keys is not overwritten; lays `.env.example` alongside if absent. `.env` is never committed, `.env.example` is committed.
0b. **Carrying over existing root content (first installation only).** A project root where working folders and files already lie is brought to order: the installer (`scripts/layout/adopt-existing.cjs`, mode `--adopt`) moves them into `project/source/`. Only the service files stay at the root: the protocol root (`.claude`, `.codex`, `.agents`, `.forma`, `.devtool`, `AGENTS.md`, `.mcp.json`), the Git files (everything on `.git*`: `.git`, `.github`, `.gitignore`, `.gitattributes`), `.env*` and `project/` itself. Everything else moves without exceptions - `README*`, `LICENSE*`, `package.json`, `node_modules`, `.vscode`, a root `CLAUDE.md` (its unique content after the move goes into `.claude/rules/claude-8.md`, step 2): at the root there must be nothing foreign that attention and tokens drain into. A taken name in `project/source/` gets a suffix (`name-2`), nothing is overwritten; an existing `project/` is not touched, the skeleton is not laid into it (step 1). The move is planned before the protocol is copied and executed after it. An update (`.forma/install-manifest.json` present) skips the step. The journal - `.forma/living/migrations/adopt-<stamp>.json`; rollback - `node <skill-dir>/../../scripts/layout/adopt-existing.cjs --rollback <journal> --root .`; a dry run without flags shows what would move. Relative paths inside what moved (builds, imports, links) are not fixed by the installer - say so to the human in the report.
1. **Two categories of goals, two different rules - not one common "ask if present".**
- **The schema's configuration** - everything from steps 2-6 and the service scripts (`.claude/rules/claude-8.md`, `.claude/agents/`, `.agents/rules/gemini-8.md`, `.agents/plugins/forma/`, `.codex/`, `.claude/hooks/*`, `.claude/scripts/*`, `manual/`; `living/` is not configuration, see step 6). It is the same for any project and is not edited by hand (`AGENTS.md` §2, "Housekeeping" - `Intent` updates these files directly, outside the production route). Therefore: **synchronize always, without questions** - no path, create it; the path exists, overwrite with the template's content as is, silently. A divergence from the template is either an older version (which the update is for), or a hand edit against the rule (also overwrite: two sources of truth for one schema are not started).
- **An installed project template outranks the core on its own files.** If `project/config/PROJECT.md`, block "Project template", records a template name (not `none`), everything that template lays into the project (its `skills/<name>/` in the plugin `templates/<template>/`, `project/config/SITE.md`, `project/config/ROUTE.md`, `project/brief/idea.md`, `.claude/scripts/` - `site.cjs`, `site-php.cjs`, `loco-pipeline.py`, `po_shift_check.py`, `tokenator_translator.py`), the core **never overwrites** - even on a name collision with a core skill. Such a path is skipped silently and named in step 9's report: "left to the template `<name>` version". Only the template's own installer may update these files, on the human's request.
- **`project/` in full** (step 7) - the specifics of the concrete project: thresholds, goals, the brief, the values, everything the human and `Intent` filled in along the way. **Do not touch at all if a `project/` directory already exists in any form** - create, overwrite, or even separately compare not a single file or subdirectory inside. The skeleton (seven files + eight directories) is created only at the very first installation, when there is no `project/` at the project root at all.
- At the end (step 9) report: what was synchronized in the configuration, whether `project/` was touched (yes - first installation only, otherwise never).
2. **Copy the common rules - three files, not one.** The protocol law (§1-7) lives as one file, the same for any environment; each engine keeps only its own §8 "Engine architecture" (where the roles lie, on which models, how the subagent is invoked, from which fields the spend is read):
- `core/AGENTS.md` → `AGENTS.md` at the project root - **the source of truth, only it is edited**;
- `adapters/claude/rules/claude-8.md` → `.claude/rules/claude-8.md` - Claude Code's §8, without an import line. **A root `CLAUDE.md` is not created**: Claude Code v2.1.277+ reads `AGENTS.md` and `.claude/rules/` itself; the engine's specifics are written only in its §8 file. A `CLAUDE.md` found at the root on an update - move its unique content into `.claude/rules/claude-8.md` and ask the human about deleting it (the human's code phrase; its literal lives in `AGENTS.md` §1); `sync-engines --check` names it a violation;
- `adapters/gemini/rules/gemini-8.md` → `.agents/rules/gemini-8.md` at the project root (step 3b) - the same line plus Antigravity's §8.
Copies of §1-7 in two wrapper files must not exist under any circumstances - that is two sources of truth for one schema. Checked mechanically: `node .claude/scripts/sync-engines.cjs --check` parses the construction of a separate "Root rules" line and falls if a wrapper lost the import, if `AGENTS.md` is absent, or if a path or field belonging to one engine has leaked into `AGENTS.md`.
All three environments read `AGENTS.md` natively: Claude Code (v2.1.277+) and Codex - as the root file, Antigravity - also through the import in `gemini-8.md`. An environment without native reading (Amazon Bedrock, telemetry off) - step 4a's fallback.
3. **Copy the five roles.** `adapters/claude/agents/*.md` → `.claude/agents/` in the project (create the directory if absent). The files are generalized - no domain details of a specific matter in them; if you find such when reading - that is a bug, report to the human, do not fix it yourself.
**No placeholders in `tools:`.** The `tools:` list in a role's frontmatter is copied as is and must work at once: a name of the form `mcp__<something>__*` with angle brackets is a template bug, not a task for the installer - in a fresh project such a name resolves into no tool. Finding one - report to the human, do not substitute a value yourself. MCP tools specific to a project (a site adapter, an own API etc.) are not part of the template core at all - `Kit` writes them in when equipping (`agents/on-demand/kit-project-kitting.md`) or a companion preset brings them for a specific environment (e.g., `forma-wordpress-novamira` - Novamira MCP).
3b. **Copy the Gemini side (Antigravity).** The same protocol, for the Google Antigravity/Gemini environment - in parallel with steps 2-3, not instead of them:
**The adapter is in beta** (`adapters/gemini/BETA.md`): the installer installs it itself (`--engines gemini`), the steps below are what exactly it lays; no need to repeat them by hand. All paths are counted from the directory of this `SKILL.md`.
- `adapters/gemini/rules/gemini-8.md` → `.agents/rules/gemini-8.md` at the project root - the Antigravity wrapper: the import `@AGENTS.md` plus its environment's §8 (step 2 above). Carries no own copy of §1-7.
- `adapters/gemini/plugin.json` → `.agents/plugins/forma/plugin.json`.
- `adapters/gemini/agents/` → `.agents/plugins/forma/agents/`, **recursively, together with `on-demand/`**. Do not use the mask `agents/*.md`: PowerShell does not capture a nested directory. Before copying create exactly the destination directory, then run: `New-Item -ItemType Directory -Force .agents/plugins/forma/agents | Out-Null; Copy-Item -Recurse -Force <skill-dir>/adapters/gemini/agents/* .agents/plugins/forma/agents/`. After that six files must exist in `.agents/plugins/forma/agents/on-demand/`.
- `adapters/gemini/mcp_config.json` does not carry MCP settings: in the package it deliberately holds an empty `mcpServers` object. Do not overwrite the project's existing configuration; concrete servers are added by `Kit` or the human after the decision about the equipment.
3c. **Copy the Codex side - only if Codex is awake (`core/engines.json`); skip for now.** **The adapter is in beta** (`adapters/codex/BETA.md`): the installer installs it itself (`--engines codex`) and runs `sync-codex --apply` if `.claude/agents` sits nearby; the steps below are what exactly it lays. The template - `adapters/codex/` in the directory of this same skill (paths counted from there):
- `adapters/codex/.codex/CODEX-8.md` → `.codex/CODEX-8.md` and `adapters/codex/.codex/CLAUDE-COMPAT.md` → `.codex/CLAUDE-COMPAT.md` - **this environment's §8** and the executable boundary with the canonical Claude roles. They carry no own copy of §1-7; the root `AGENTS.md` stays the source.
- `adapters/codex/.codex/config.toml` → `.codex/config.toml` - **except the `[mcp_servers.*]` sections**: MCP is always project-specific, show it to the human and leave the filling to `Kit`/the human, in the same order as `mcp_config.json` (3b) and `.mcp.json` (the Claude side).
- `adapters/codex/.codex/agents/*.toml` → `.codex/agents/` - the five subagent definitions and the mechanical `extractor`.
- `adapters/codex/.codex/roles/*.md` → `.codex/roles/` - the five roles' adapters and the `extractor`. **Both kinds are obligatory:** the TOML says what the agent runs as, the Markdown directs to the full canonical Claude body; do not copy a role as a second meaningful text.
- `adapters/codex/.codex/roles/on-demand/*.md` → `.codex/roles/on-demand/` - the "on demand" layer. **Do not skip it:** the law references these files by bare name (§2, §5 and on), and without them a Codex node will not find the instruction it was sent for. These are short adapters leading to the canonical Claude procedures; `sync-engines.cjs` checks their link and structure, not a byte-for-byte match of the text.
- `adapters/codex/.codex/hooks.json` → `.codex/hooks.json`; `adapters/codex/.codex/hooks/*.ps1` → `.codex/hooks/` - start checks, the delete guard and logging, on PowerShell. **This is not "unfinished Bash":** an edit of the shell hooks is carried here by meaning, not by copying.
- `adapters/codex/.codex/tests/*.ps1` → `.codex/tests/`.
- Codex's project skills are discovered from `.agents/skills/`. After `.codex/` and the canonical `.claude/skills/` are installed, run `node .codex/scripts/sync-codex.cjs --apply`, then `node .codex/scripts/sync-codex.cjs --check`. The script mirrors **every** directory `.claude/skills/<name>/` together with all nested files and updates the manifest. Do not copy this set selectively by hand: `verify-parity.ps1` checks the dynamic composition and all files. MCP sections are not part of the template and are added only after the human's decision.
**Two differences of this environment that must not be "fixed".** The ceiling of parallel threads - three (with the main session that is all four slots of the environment); raising it is not advised. The spend per §3 is not measured here - the configuration gives the agent no token and cache-read fields, and the law answers directly: **do not estimate and do not write zeros**, but record that the fields were unavailable.
4. **Copy the hooks.** `adapters/claude/hooks/guard-delete.sh` → `.claude/hooks/guard-delete.sh`; `adapters/claude/hooks/check-ready.sh` → `.claude/hooks/check-ready.sh`; `adapters/claude/hooks/check-plugin-update.sh` → `.claude/hooks/check-plugin-update.sh`; `adapters/claude/hooks/check-dashboard.sh` → `.claude/hooks/check-dashboard.sh`; `adapters/claude/hooks/tool-usage.sh` → `.claude/hooks/tool-usage.sh`; `adapters/claude/hooks/intent-start.sh` → `.claude/hooks/intent-start.sh`; `adapters/claude/hooks/check-card.sh` → `.claude/hooks/check-card.sh`; `adapters/claude/hooks/check-clean-zone.sh` → `.claude/hooks/check-clean-zone.sh`. On a Unix-like system set the executable bit (`chmod +x`) on all eight. `intent-start.sh` is also a `SessionStart` (`startup`): it feeds the main session the `Intent` role and `intent-session-start.md`, shows the human the link to the dashboard (`AGENTS.md` §1). `check-card.sh` and `check-clean-zone.sh` - `PostToolUse` with the matcher `Write|Edit`, two entries in one matcher: the first checks the board card at the moment of writing, the second reports a project trace in a clean zone (prohibition 16). `guard-delete.sh` and `tool-usage.sh` connect through `PreToolUse`; `check-ready.sh`, `check-plugin-update.sh` and `check-dashboard.sh` - through `SessionStart` (matcher `startup`, for `check-plugin-update.sh` a `timeout` of 15) - in the project's `.claude/settings.json`, all three as separate entries in one matcher, their order irrelevant. `check-dashboard.sh` keeps the dashboard self-healing (restarts `dashboard/serve.js` when it does not answer); `tool-usage.sh` writes the log of tool calls for the "Deck" tab. `check-plugin-update.sh` says when a plugin version newer than the installed one has appeared on GitHub and stays silent in every other case - including no `gh` or no network. **The protocol has no "update" command deliberately:** a command must be remembered, while a hook fires by itself; the update is requested in words in a live conversation, and the `forma` skill responds to them per its description. If the settings file does not exist yet, create it with these entries; if it exists, add them, not replacing the rest.
4a. **The fallback for feeding §8.** `adapters/claude/hooks/load-engine-section.sh` → `.claude/hooks/load-engine-section.sh`, **do not register**. The native reading of `.claude/rules/` is verified by fact on v2.1.278 but is not documented: if it disappears, the hook's entry in `.claude/settings.json` (`SessionStart`, matcher `startup|resume|clear`) returns §8 with one edit. While the reading works, registering would deliver §8 twice.
4b. **Copy the service scripts.** `adapters/claude/scripts/*` → `.claude/scripts/` in the project: `sync-engines.cjs` for 100% parity of the two engines, `external-model-bridge.cjs` for the bridge to an external model, `build-done-cards-graph.cjs` for building the experience graph over closed cards, `session-economy.cjs` for measuring re-reading against work over the engine's transcripts, `tool-usage.cjs` for parsing the log of the `tool-usage.sh` hook.
4c. **The interviews and the board are the core; the adapter delivers them.** First the core: `core/skills/` → `skills/` at the project root (both interviews - the source), `core/board/` → `board/` (`new-card.cjs`, `check-board.cjs`, `DATA-MODEL.md`), `core/.devtool/features/` → `.devtool/features/` (the board's blank; on an update the cards are not touched - only carry the folder over if it does not exist). Then the Claude adapter's delivery: `skills/grilling/` → `.claude/skills/grilling/` (the terminal) and `skills/forma-grill-with-ui/` → `.claude/skills/forma-grill-with-ui/` (the browser page: questions with recommendations, answers in any order, a discussion per each question separately) - by copy from the core, without edits; `sync-engines --check` names a copy's divergence from the core. The hook `check-card.sh` (step 4) calls `node board/check-board.cjs <card>`; cards are created by `node board/new-card.cjs`. **Both interviews are part of the core, not companions**: epic "4. Project resources" is one of the four process ones, and `agents/on-demand/intent-goal-opening.md` references both files by name and chooses between them itself - `forma-grill-with-ui` by default when a browser is available, else `grilling`; the choice rule - there too, "Epic". If you fail to copy one of the two - a fresh project has a broken link from day one.
4d. **Copy the dashboard and the knowledge graph.** `core/dashboard/` → the directory `dashboard/` at the project root (not into `.devtool/` - that belongs to the Kanban Markdown extension). First create the destination directory, then copy its content: `New-Item -ItemType Directory -Force dashboard | Out-Null; Copy-Item -Recurse -Force <skill-dir>/core/dashboard/* dashboard/`, then the Claude adapter's part - the readers of its transcripts: `Copy-Item -Force <skill-dir>/adapters/claude/dashboard/* dashboard/`. Do not copy a single file onto the `dashboard` path: that creates a file instead of a directory. After copying `dashboard` must be a directory with `serve.js` and `index.html`. Started by the human manually: `node dashboard/serve.js` → `http://localhost:5050/`. Runtime artifacts (`data.json`, `watch.log`, `graphify-pending.json`) are written into `dashboard/.cache/` - add this path to the project's `.gitignore` if it is not there yet. The knowledge graph is not a separate window but tabs of the same dashboard ("Graph: Project"/"Graph: Engine"), fed by a forked copy of the `graphify` skill: `adapters/claude/skills/graphify/` → `.claude/skills/graphify/` in the project - an own copy pinned by version, not a dependency on a global install (if the project already has a global `~/.claude/skills/graphify`, there is no conflict - Claude Code takes the project skill of the same name over the global one, but say explicitly to the human that the project copy is the one used).
**Set up `dashboard/graphify.config.json` together with the human, not for them.** `trigger.mode` - `"off"` (the template default; the graph is rebuilt only by an explicit `/graphify` command) or `"history-accumulation"` (`trigger.threshold`, default `20`); `executor.channel` - `"internal"` (default, through `Agent` in a live session) or `"external"` (no live session, through `external-model-bridge.cjs` - needs a key in `.env` and spends money, only on the human's explicit request); `budget.maxUsdPerBuild`/`alertUsd` make sense only at `channel: "external"`. Do not choose yourself - this is a decision about money and automation.
5. **Copy no commands - there are none.** The protocol creates no `.claude/commands/`: neither "tell me what to do now", nor "update". A command must be remembered, and that is exactly what does not work - an installed plugin once fell behind by dozens of versions because there was no one to remind about calling `/forma-update`. Instead: the session-start hook speaks of the update (step 4), and the update itself is requested in words - the `forma` skill responds to "update forma" per its description and does exactly what the command did. The direction of work is set by the law (`AGENTS.md`) and the state of the files, in a live conversation.
6. **Copy the schema documentation and the living catalog.** `core/manual/` → `manual/` at the project root (`en/` - the original, `ru/` - an exact copy, nine sections in each, `01-analytics/` … `09-results/`, the engines' complete references in `03-protocol/`; `README.md`; images - `en/assets/`, `ru/assets/`; not translated to the project language) - only the explanation of the system, universal, not edited for a specific matter (`manual/en/03-protocol/SCHEME.md` inside, "Two halves"). `core/living/` → `living/` (`CHANGELOG.md`, `dictionary/`; `living/graphs/` is created by the graph scripts) - everything that grows and is rebuilt. `living/` on an update **is not overwritten**: what is missing - carry it over, what exists - leave, it is already the project's history.
7. **Copy the project structure's skeleton.** `core/project/` → `project/` at the project root - seven files (`CONFIG.md`, `PROJECT.md`, `ROADMAP.md`, `ROUTE.md`, `JOURNAL.md`, `SETUP.md`, `VALUE.md`) and eight directories, each with a blank `README.md`: `VARS/`, `brief/` (plus the skeletons `brief/idea.md` - seven sections of the idea, the interview runs over it, and `brief/landing-prompt.md` - the landing page's prompt, written during the interview; a project template with its own `idea.md` lays it on top), `cards/`, `docs/`, `experience/`, `goals/`, `mockups/`, `reference/`. The skeleton - headings, tables, record templates, the method in each directory's `README.md`; the content empty, filled along the way (the thresholds and equipment - in `PROJECT.md`, by hand, before the start; the goals - in `ROADMAP.md`, per `SETUP.md`; the files inside the directories themselves - `brief/interview.md`, `goals/goal-NN/GOAL.md` etc. - as preparation and goals proceed, the directories already in place).
8. **One detachable remains - the project template.** The interviews (both variants) and the dashboard with the graph are already in - part of the core, steps 4c-4d, no second installation needed. The only thing left to ask the human separately is the project template; the kanban board is obligatory, but it is an external editor extension, not a plugin of this marketplace (step 9).
**The project template - not a default but the human's choice.** Name the fork and let the human decide, do not decide for them:
- **take a ready template** - the route is already written and, if it is `a proven ready one`, walked to the result. Then the interview's fifth position reduces to a check. What this plugin has - `templates/` (e.g., `forma-wordpress-novamira` for a WordPress/Elementor/Voxel site through Novamira MCP);
- **develop your own from scratch** - "an unknown zone". No route exists, `project/config/ROUTE.md` stays a blank, and the route is developed by `Intent` with the human **at the moment of the interview**, as the fifth position alongside the other four. It is then refined along the way - that is normal, not a sign of bad planning.
Whatever the human chose, **write the status into `project/config/PROJECT.md`, block "Project template"**: by it `Intent` picks the branch instead of guessing by the presence of files. Chose nothing now - leave `none` and say the field is edited at any moment.
8b. **Run the integrity check and show the human the result:** `node .forma/verify/verify-install.cjs` (the installer already ran it at the end; the separate command - for checking a live project at any moment). Three layers: the inventory `.forma/install-manifest.json` (sha256 of the installer's files, the `project/` skeleton - the count of headings, not a hash: the human's zones get filled), the closure (every `.cjs` parses, every `.json` reads, hooks ↔ files ↔ `settings.json`, references by bare name from `AGENTS.md`/§8 lead to files, no `mcp__<…>__*` in `tools:`, no root `CLAUDE.md`), behavior (the start hook and `sync-engines --check` answer, the model is not invoked). The answer - `✓ install is intact` or a list "address - expected - found". Red at the main engine - **stop** (exit 1): fix per the list and repeat; at the secondary ones - a warning. An engine without a ready adapter (Codex, Gemini) answers "not installed" until its adapter has a reconciliation profile. Inside the behavioral layer runs the former core reconciliation: `node .claude/scripts/sync-engines.cjs --check`. The expected answer - `OK: AGENTS.md - the single law, the core is whole, the engine adapters passed the conformance test.` The reconciliation checks the core's wholeness (the law, `manual/…/ECONOMY.md`, the economy module and its test) and every installed adapter by its conformance test; it does not copy roles between engines, `--apply` writes nothing.
**Running it here, not "sometime later", is deliberate.** This reconciliation checks not only the parity of the two environments but also the board's epics: it takes the process names from `PROJECT.md`, the goals' existence from the directory names of `project/goals/`, and compares the substantive epics' names with `ROADMAP.md`. In a freshly installed project there is no board and no goals - the reconciliation accounts for that and stays silent. The first run is needed here for another thing: the human sees the instrument at work at once, under supervision, instead of meeting it a month later mid-work.
The answer not `OK` - take it apart and report to the human, without moving to step 9: on a fresh installation a divergence means a file not carried over, not a human error.
9. **Report and show what is next.** A summary per step 1: what was newly carried over this round, what already stood and stayed untouched, what required a human decision. Freshly laid files are empty: the thresholds in `PROJECT.md` are not set, the goals in `ROADMAP.md` do not exist (if this is the first installation, not carrying over what is missing into an already prepared project). **At the first installation - explicitly warn about the obligatory external dependency** (`manual/en/03-protocol/ZONES.md`, zone 3 "Core - the kanban board"): without the VS Code extension Kanban Markdown (`code --install-extension LachyFS.kanban-markdown`) the cards on the board remain just a folder of `.md` files - not cosmetics: without it the dashboard/graph/criterion checks have no data source. The next step for a first installation - preparation per `project/config/SETUP.md`, starting with the interview; that is already `Intent`'s work with the human, not this skill's.
## What this skill does not do
- it does not invent the content of `PROJECT.md`/`ROADMAP.md`/goals - the skeleton is empty, filling it is `Intent`'s work in conversation with the human, not the installation's;
- on a repeated call **it does not touch `project/` at all** if it already exists - neither at the file level nor at the directory level; only the schema's configuration is subject to synchronization (step 1);
- it creates the directories (`brief/`, `VARS/`, `docs/`, `goals/`, `mockups/`, `reference/`) empty, with a single `README.md` in each, only at the first installation - the files inside them (`interview.md`, `goal-NN/GOAL.md` etc.) are started as preparation (`SETUP.md`) and goals proceed, not at the installation;
- it publishes and commits nothing itself - it only lays files on disk; the commit/publication is done by the user in the same order as any other edit of theirs;
- it adds no MCP servers for the human: `adapters/gemini/mcp_config.json` ships empty, and the project's existing configuration is not overwritten.
Pliki w paczce (719)
- SKILL.md 34 KB
- adapters/claude/forma-adapter.cjs 5 KB
- adapters/claude/permission-levels.json 3 KB
- adapters/claude/verify-profile.cjs 7 KB
- adapters/claude/agents/asset-describer-haiku.md 1 KB
- adapters/claude/agents/core.md 19 KB
- adapters/claude/agents/extractor.md 1 KB
- adapters/claude/agents/intent.md 29 KB
- adapters/claude/agents/kit.md 38 KB
- adapters/claude/agents/plumb-haiku.md 2 KB
- adapters/claude/agents/run.md 15 KB
- adapters/claude/agents/spec.md 18 KB
- adapters/claude/agents/on-demand/card-and-board.md 20 KB
- adapters/claude/agents/on-demand/citing-schema-to-human.md 1 KB
- adapters/claude/agents/on-demand/degraded-mode.md 1 KB
- adapters/claude/agents/on-demand/intent-cycle-closing.md 3 KB
- adapters/claude/agents/on-demand/intent-demo.md 4 KB
- adapters/claude/agents/on-demand/intent-documentation.md 2 KB
- adapters/claude/agents/on-demand/intent-emergency.md 2 KB
- adapters/claude/agents/on-demand/intent-goal-opening.md 29 KB
- adapters/claude/agents/on-demand/intent-housekeeping.md 13 KB
- adapters/claude/agents/on-demand/intent-session-start.md 8 KB
- adapters/claude/agents/on-demand/intent-visual-check.md 2 KB
- adapters/claude/agents/on-demand/kit-arsenal.md 3 KB
- adapters/claude/agents/on-demand/kit-experience.md 1 KB
- adapters/claude/agents/on-demand/kit-graphs.md 2 KB
- adapters/claude/agents/on-demand/kit-limits.md 2 KB
- adapters/claude/agents/on-demand/kit-project-kitting.md 3 KB
- adapters/claude/agents/on-demand/kit-recon.md 3 KB
- adapters/claude/agents/on-demand/kit-route8.md 1 KB
- adapters/claude/agents/on-demand/route-choice.md 11 KB
- adapters/claude/agents/on-demand/run-external-model.md 1 KB
- adapters/claude/agents/on-demand/spec-cycle-boundary.md 1 KB
- adapters/claude/agents/on-demand/spec-route8.md 2 KB
- adapters/claude/agents/on-demand/spec-slicing-details.md 6 KB
- adapters/claude/agents/on-demand/spec-statistics.md 1 KB
- adapters/claude/agents/on-demand/spec-trial-pilot.md 3 KB
- adapters/claude/agents/on-demand/spec-unattainable-value.md 1 KB
- adapters/claude/agents/on-demand/spend-line.md 11 KB
- adapters/claude/agents/run/run-dashboard-schema.md 4 KB
- adapters/claude/agents/run/run-image-series.md 4 KB
- adapters/claude/agents/run/run-image.md 3 KB
- adapters/claude/agents/run/run-mechanical.md 1 KB
- adapters/claude/agents/run/run-protocol-adapters.md 5 KB
- adapters/claude/agents/run/run-protocol-anglicize.md 4 KB
- adapters/claude/agents/run/run-protocol-audit.md 3 KB
- adapters/claude/agents/run/run-protocol-board.md 5 KB
- adapters/claude/agents/run/run-protocol-core.md 5 KB
- adapters/claude/agents/run/run-protocol-dashboard.md 4 KB
- adapters/claude/agents/run/run-protocol-docs.md 3 KB
- adapters/claude/agents/run/run-protocol-driver.md 4 KB
- adapters/claude/agents/run/run-protocol-engine.md 4 KB
- adapters/claude/agents/run/run-protocol-script.md 3 KB
- adapters/claude/agents/run/run-protocol-summary.md 2 KB
- adapters/claude/agents/run/run-protocol-testfix.md 4 KB
- adapters/claude/agents/run/run-python.md 1 KB
- adapters/claude/agents/run/run-resources-registry.md 3 KB
- adapters/claude/agents/run/run-schema-markup.md 2 KB
- adapters/claude/agents/run/run-script.md 2 KB
- adapters/claude/agents/run/run-site-build.md 1 KB
- adapters/claude/agents/run/run-site-php.md 3 KB
- adapters/claude/agents/run/run-text.md 1 KB
- adapters/claude/agents/run/run-vector.md 1 KB
- adapters/claude/agents/run/run-visual-check.md 1 KB
- adapters/claude/dashboard/floor.cjs 10 KB
- adapters/claude/dashboard/subagent-transcript.cjs 10 KB
- adapters/claude/dashboard/subagent-transcript.test.cjs 1 KB
- adapters/claude/dashboard/subagents-log.cjs 18 KB
- adapters/claude/hooks/auto-graphs.sh 1 KB
- adapters/claude/hooks/check-card.sh 1 KB
- adapters/claude/hooks/check-clean-zone.sh 1 KB
- adapters/claude/hooks/check-credentials.sh 1 KB
- adapters/claude/hooks/check-dashboard.sh 1 KB
- adapters/claude/hooks/check-plugin-update.sh 3 KB
- adapters/claude/hooks/check-ready.sh 1 KB
- adapters/claude/hooks/guard-delete.sh 9 KB
- adapters/claude/hooks/intent-start.sh 2 KB
- adapters/claude/hooks/load-engine-section.sh 1 KB
- adapters/claude/hooks/sync-extensions.sh 1 KB
- adapters/claude/hooks/tool-usage.sh 4 KB
- adapters/claude/hooks/unlock-delete.sh 1 KB
- adapters/claude/rules/card-reminders.md 1 KB
- adapters/claude/rules/claude-8.md 9 KB
- adapters/claude/scripts/auto-graphs.cjs 3 KB
- adapters/claude/scripts/build-engine-graph-html.cjs 3 KB
- adapters/claude/scripts/build-engine-graph.cjs 11 KB
- adapters/claude/scripts/build-graph-html.cjs 4 KB
- adapters/claude/scripts/build-kit-graphs.cjs 52 KB
- adapters/claude/scripts/build-site-graph.cjs 16 KB
- adapters/claude/scripts/build-tendons.cjs 15 KB
- adapters/claude/scripts/card-session-spend.cjs 18 KB
- adapters/claude/scripts/card-session-spend.test.cjs 11 KB
- adapters/claude/scripts/card-subagent-spend.cjs 5 KB
- adapters/claude/scripts/card-subagent-spend.test.cjs 9 KB
- adapters/claude/scripts/claude-economy.cjs 12 KB
- adapters/claude/scripts/claude-economy.test.cjs 6 KB
- adapters/claude/scripts/cycle-status.cjs 4 KB
- adapters/claude/scripts/dashboard-view.cjs 8 KB
- adapters/claude/scripts/dispatch-prompt.cjs 7 KB
- adapters/claude/scripts/engine-switch.cjs 6 KB
- adapters/claude/scripts/external-model-bridge.cjs 13 KB
- adapters/claude/scripts/forma-adapter.test.cjs 8 KB
- adapters/claude/scripts/generate-diffui-focus-assets.cjs 7 KB
- adapters/claude/scripts/graph-query.cjs 9 KB
- adapters/claude/scripts/instruction-weight.cjs 5 KB
- adapters/claude/scripts/ledger-recorder.cjs 15 KB
- adapters/claude/scripts/ledger-recorder.test.cjs 12 KB
- adapters/claude/scripts/permissions-level.cjs 5 KB
- adapters/claude/scripts/permissions-level.test.cjs 3 KB
- adapters/claude/scripts/plumb-measure.cjs 4 KB
- adapters/claude/scripts/plumb-run.cjs 1 KB
- adapters/claude/scripts/plumb-run.test.cjs 4 KB
- adapters/claude/scripts/plumb-scrub.cjs 7 KB
- adapters/claude/scripts/plumb-scrub.test.cjs 4 KB
- adapters/claude/scripts/plumb-switch.cjs 2 KB
- adapters/claude/scripts/resources.cjs 31 KB
- adapters/claude/scripts/session-economy.cjs 24 KB
- adapters/claude/scripts/subagent-spend.cjs 3 KB
- adapters/claude/scripts/subagent-spend.test.cjs 3 KB
- adapters/claude/scripts/sync-engines.cjs 9 KB
- adapters/claude/scripts/test-ledger-parity-reverse.cjs 6 KB
- adapters/claude/scripts/tool-usage.cjs 14 KB
- adapters/claude/scripts/checks/adapter-conformance.cjs 3 KB
- adapters/claude/scripts/checks/claude-delivery.cjs 2 KB
- adapters/claude/scripts/checks/codex.cjs 8 KB
- adapters/claude/scripts/checks/engine-impersonal.cjs 2 KB
- adapters/claude/scripts/checks/index.cjs 1 KB
- adapters/claude/scripts/checks/kimi.cjs 3 KB
- adapters/claude/scripts/checks/law-pointers.cjs 2 KB
- adapters/claude/scripts/checks/root-rules.cjs 5 KB
- adapters/claude/scripts/checks/rules-size.cjs 1 KB
- adapters/claude/scripts/checks/support/clean-zone.cjs 4 KB
- adapters/claude/scripts/checks/support/dormant.cjs 1 KB
- adapters/claude/scripts/checks/support/engine-roots.cjs 2 KB
- adapters/claude/scripts/checks/support/gen-gemini-roles.cjs 9 KB
- adapters/claude/scripts/checks/support/role-parity.cjs 14 KB
- adapters/claude/scripts/checks/support/test-engine-roots.cjs 5 KB
- adapters/claude/scripts/checks/support/test-gen-gemini-roles.cjs 4 KB
- adapters/claude/scripts/checks/support/test-law-pointers.cjs 1 KB
- adapters/claude/scripts/checks/support/test-role-parity.cjs 3 KB
- adapters/codex/BETA.md 1 KB
- adapters/codex/README.md 1 KB
- adapters/codex/.agents/skills/engine-switch/SKILL.md 2 KB
- adapters/codex/.agents/skills/excalidraw-diagrams/SKILL.md 5 KB
- adapters/codex/.agents/skills/graph-build/SKILL.md 6 KB
- adapters/codex/.agents/skills/graphify/.graphify_version 1 KB
- adapters/codex/.agents/skills/graphify/SKILL.md 50 KB
- adapters/codex/.agents/skills/human-register/SKILL.md 6 KB
- adapters/codex/.agents/skills/law-shrink/SKILL.md 3 KB
- adapters/codex/.agents/skills/project-knowledge/SKILL.md 4 KB
- adapters/codex/.agents/skills/skill-authoring/SKILL.md 8 KB
- adapters/codex/.agents/skills/spend-parity/SKILL.md 3 KB
- adapters/codex/.codex/CHANGELOG.md 1 KB
- adapters/codex/.codex/CLAUDE-COMPAT.md 4 KB
- adapters/codex/.codex/CODEX-8.md 17 KB
- adapters/codex/.codex/README.md 3 KB
- adapters/codex/.codex/config.toml 1 KB
- adapters/codex/.codex/forma-adapter.cjs 1 KB
- adapters/codex/.codex/generated-skill-mirror.json 1 KB
- adapters/codex/.codex/hooks.json 2 KB
- adapters/codex/.codex/agents/core.toml 1 KB
- adapters/codex/.codex/agents/extractor.toml 1 KB
- adapters/codex/.codex/agents/intent.toml 1 KB
- adapters/codex/.codex/agents/kit.toml 1 KB
- adapters/codex/.codex/agents/plumb.toml 1 KB
- adapters/codex/.codex/agents/run-image-series.toml 1 KB
- adapters/codex/.codex/agents/run-image.toml 1 KB
- adapters/codex/.codex/agents/run-mechanical.toml 1 KB
- adapters/codex/.codex/agents/run-python.toml 1 KB
- adapters/codex/.codex/agents/run-schema-markup.toml 1 KB
- adapters/codex/.codex/agents/run-script.toml 1 KB
- adapters/codex/.codex/agents/run-site-build.toml 1 KB
- adapters/codex/.codex/agents/run-site-php.toml 1 KB
- adapters/codex/.codex/agents/run-text.toml 1 KB
- adapters/codex/.codex/agents/run-vector.toml 1 KB
- adapters/codex/.codex/agents/run-visual-check.toml 1 KB
- adapters/codex/.codex/agents/run.toml 1 KB
- adapters/codex/.codex/agents/spec.toml 1 KB
- adapters/codex/.codex/hooks/check-card.ps1 1 KB
- adapters/codex/.codex/hooks/check-dashboard.ps1 1 KB
- adapters/codex/.codex/hooks/check-plugin-update.ps1 3 KB
- adapters/codex/.codex/hooks/check-ready.ps1 2 KB
- adapters/codex/.codex/hooks/guard-delete.ps1 2 KB
- adapters/codex/.codex/hooks/intent-start.ps1 1 KB
- adapters/codex/.codex/hooks/tool-usage.ps1 2 KB
- adapters/codex/.codex/roles/core.md 1 KB
- adapters/codex/.codex/roles/extractor.md 1 KB
- adapters/codex/.codex/roles/intent.md 1 KB
- adapters/codex/.codex/roles/kit.md 2 KB
- adapters/codex/.codex/roles/plumb.md 2 KB
- adapters/codex/.codex/roles/run.md 1 KB
- adapters/codex/.codex/roles/spec.md 1 KB
- adapters/codex/.codex/roles/on-demand/citing-schema-to-human.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-cycle-closing.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-demo.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-documentation.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-emergency.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-goal-opening.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-housekeeping.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-session-start.md 1 KB
- adapters/codex/.codex/roles/on-demand/intent-visual-check.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-arsenal.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-experience.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-graphs.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-limits.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-project-kitting.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-recon.md 1 KB
- adapters/codex/.codex/roles/on-demand/kit-route8.md 1 KB
- adapters/codex/.codex/roles/on-demand/route-choice.md 1 KB
- adapters/codex/.codex/roles/on-demand/run-external-model.md 1 KB
- adapters/codex/.codex/roles/on-demand/spec-cycle-boundary.md 1 KB
- adapters/codex/.codex/roles/on-demand/spec-route8.md 1 KB
- adapters/codex/.codex/roles/on-demand/spec-slicing-details.md 1 KB
- adapters/codex/.codex/roles/on-demand/spec-statistics.md 1 KB
- adapters/codex/.codex/roles/on-demand/spec-trial-pilot.md 1 KB
- adapters/codex/.codex/roles/on-demand/spec-unattainable-value.md 1 KB
- adapters/codex/.codex/roles/on-demand/spend-line.md 1 KB
- adapters/codex/.codex/roles/run/run-image-series.md 1 KB
- adapters/codex/.codex/roles/run/run-image.md 1 KB
- adapters/codex/.codex/roles/run/run-mechanical.md 1 KB
- adapters/codex/.codex/roles/run/run-python.md 1 KB
- adapters/codex/.codex/roles/run/run-schema-markup.md 1 KB
- adapters/codex/.codex/roles/run/run-script.md 1 KB
- adapters/codex/.codex/roles/run/run-site-build.md 1 KB
- adapters/codex/.codex/roles/run/run-site-php.md 1 KB
- adapters/codex/.codex/roles/run/run-text.md 1 KB
- adapters/codex/.codex/roles/run/run-vector.md 1 KB
- adapters/codex/.codex/roles/run/run-visual-check.md 1 KB
- adapters/codex/.codex/scripts/codex-hooks.cjs 4 KB
- adapters/codex/.codex/scripts/codex-usage.cjs 11 KB
- adapters/codex/.codex/scripts/sync-codex.cjs 20 KB
- adapters/codex/.codex/tests/test-codex-usage.cjs 5 KB
- adapters/codex/.codex/tests/test-plumb-codex.cjs 7 KB
- adapters/codex/.codex/tests/test-sync-codex.cjs 16 KB
- adapters/codex/.codex/tests/verify-config.ps1 7 KB
- adapters/codex/.codex/tests/verify-parity.ps1 4 KB
- adapters/gemini/BETA.md 7 KB
- adapters/gemini/README.md 2 KB
- adapters/gemini/forma-adapter.cjs 11 KB
- adapters/gemini/hooks.json 1 KB
- adapters/gemini/mcp_config.json 1 KB
- adapters/gemini/plugin.json 1 KB
- adapters/gemini/verify-profile.cjs 7 KB
- adapters/gemini/.ledger/.gitkeep 1 KB
- adapters/gemini/agents/asset-describer-gemini.md 1 KB
- adapters/gemini/agents/asset-describer-haiku.md 1 KB
- adapters/gemini/agents/core.md 19 KB
- adapters/gemini/agents/extractor.md 1 KB
- adapters/gemini/agents/intent.md 28 KB
- adapters/gemini/agents/kit.md 38 KB
- adapters/gemini/agents/plumb-gemini.md 2 KB
- adapters/gemini/agents/run.md 15 KB
- adapters/gemini/agents/spec.md 18 KB
- adapters/gemini/agents/on-demand/card-and-board.md 20 KB
- adapters/gemini/agents/on-demand/citing-schema-to-human.md 1 KB
- adapters/gemini/agents/on-demand/degraded-mode.md 1 KB
- adapters/gemini/agents/on-demand/intent-cycle-closing.md 3 KB
- adapters/gemini/agents/on-demand/intent-demo.md 4 KB
- adapters/gemini/agents/on-demand/intent-documentation.md 2 KB
- adapters/gemini/agents/on-demand/intent-emergency.md 2 KB
- adapters/gemini/agents/on-demand/intent-goal-opening.md 29 KB
- adapters/gemini/agents/on-demand/intent-housekeeping.md 13 KB
- adapters/gemini/agents/on-demand/intent-session-start.md 8 KB
- adapters/gemini/agents/on-demand/intent-visual-check.md 2 KB
- adapters/gemini/agents/on-demand/kit-arsenal.md 3 KB
- adapters/gemini/agents/on-demand/kit-experience.md 1 KB
- adapters/gemini/agents/on-demand/kit-graphs.md 2 KB
- adapters/gemini/agents/on-demand/kit-limits.md 2 KB
- adapters/gemini/agents/on-demand/kit-project-kitting.md 3 KB
- adapters/gemini/agents/on-demand/kit-recon.md 3 KB
- adapters/gemini/agents/on-demand/kit-route8.md 1 KB
- adapters/gemini/agents/on-demand/route-choice.md 11 KB
- adapters/gemini/agents/on-demand/run-external-model.md 1 KB
- adapters/gemini/agents/on-demand/spec-cycle-boundary.md 1 KB
- adapters/gemini/agents/on-demand/spec-route8.md 2 KB
- adapters/gemini/agents/on-demand/spec-slicing-details.md 6 KB
- adapters/gemini/agents/on-demand/spec-statistics.md 1 KB
- adapters/gemini/agents/on-demand/spec-trial-pilot.md 3 KB
- adapters/gemini/agents/on-demand/spec-unattainable-value.md 1 KB
- adapters/gemini/agents/on-demand/spend-line.md 11 KB
- adapters/gemini/agents/run/run-dashboard-schema.md 5 KB
- adapters/gemini/agents/run/run-image-series.md 3 KB
- adapters/gemini/agents/run/run-image.md 3 KB
- adapters/gemini/agents/run/run-mechanical.md 1 KB
- adapters/gemini/agents/run/run-protocol-adapters.md 5 KB
- adapters/gemini/agents/run/run-protocol-anglicize.md 4 KB
- adapters/gemini/agents/run/run-protocol-audit.md 3 KB
- adapters/gemini/agents/run/run-protocol-board.md 5 KB
- adapters/gemini/agents/run/run-protocol-core.md 5 KB
- adapters/gemini/agents/run/run-protocol-dashboard.md 4 KB
- adapters/gemini/agents/run/run-protocol-docs.md 3 KB
- adapters/gemini/agents/run/run-protocol-driver.md 4 KB
- adapters/gemini/agents/run/run-protocol-engine.md 3 KB
- adapters/gemini/agents/run/run-protocol-script.md 3 KB
- adapters/gemini/agents/run/run-protocol-summary.md 2 KB
- adapters/gemini/agents/run/run-protocol-testfix.md 4 KB
- adapters/gemini/agents/run/run-python.md 1 KB
- adapters/gemini/agents/run/run-resources-registry.md 3 KB
- adapters/gemini/agents/run/run-schema-markup.md 2 KB
- adapters/gemini/agents/run/run-script.md 2 KB
Pominięte w paczce: 57 (zbyt duże albo wykonywalne), są w repozytorium.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · FormaMindPro/formamind | 100 | MIT |
| figueirego/forma | 100 | MIT |
| figueirego/forma | 100 | MIT |
| figueirego/forma | 100 | MIT |
| figueirego/forma | 100 | MIT |
| Pukhaan/forma-board | 100 | AGPL-3.0 |
| namzug16/forma | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki d2f6df5564ca (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100forma-templating
Pakuje bieżący projekt Forma jako instalowalny szablon pod nazwą podaną przez człowieka, tylko na jego wyraźne żądanie. - 100setup
Instaluje, konfiguruje, aktualizuje lub usuwa t3-orchestration w projekcie w T3 Code, gdy użytkownik o to prosi. - 100forma-reference
Szuka przykładów w katalogu Forma, podaje autorów materiałów i porównuje różnice wizualne i techniczne na podstawie kodu i działania. - 100forma-reference
Szuka przykładów w katalogu Forma, podaje autorów materiałów i porównuje różnice wizualne i techniczne na podstawie kodu i działania. - 100formae-apply
Wdraża infrastrukturę z pliku forma, uzgadnia lub aktualizuje stack i stosuje zaplanowane zmiany infrastruktury w formae. - 90sb:update
Sprawdza najnowszą wersję Silver Bullet na GitHubie, pokazuje zmiany od zainstalowanej wersji i instaluje aktualizację po potwierdzeniu.