ai-dlc-sdk-angular
In-repository AI-DLC SDK for Angular embedded front-panel GUIs: GitHub Copilot CLI implements screens from HTML/CSS + OpenAPI + ui-bindings, loops (max N) until Excel-derived subsystem tests pass, then reverse-generates design docs and benchmarks QCD.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Check it deeper on your machine
It has changed since its scan on 19 Nov 2026.
As a Claude Code plugin, install it and then ask Claude Code:
/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/sun-flat-yamada/ai-dlc-sdk-angular with scanskillsafety
Or from a terminal, clone it and run it:
git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/sun-flat-yamada/ai-dlc-sdk-angular'
It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.
Badge for your README
Markdown
[](/listings/sun-flat-yamada--ai-dlc-sdk-angular)HTML
<a href="/listings/sun-flat-yamada--ai-dlc-sdk-angular"><img src="/api/listings/sun-flat-yamada--ai-dlc-sdk-angular/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Everything runs in the bundled **sandbox** ([docs/sandbox.md](docs/sandbox.md)). The host needs only Docker and Node.js — no `npm ci` on the host. In VS Code, **Reopen in Container** does the same.
npm run sandbox -- up # build + start the sandbox (firewall, npm ci, Copilot CLI, Chromium) npm run sandbox -- shell # a shell inside the sandbox; the commands below run there npm run aidlc:parse-excel # Excel -> .ai-dlc/outputs/generated-tests/subsystem-cases.json npm run aidlc:validate-inputs # Gate 0 + binding matrix (.ai-dlc/outputs/reports/binding-matrix.md) npm run aidlc:init-check -- --dry-run npm run aidlc:run -- --max-iterations 3 --dry-run # whole pipeline without calling Copilot
Real run (needs a [GitHub Copilot](https://docs.github.com/copilot/how-tos/copilot-cli) subscription and a clean git tree). Copilot CLI is already installed in the sandbox:
copilot login # once, inside the sandbox (or COPILOT_GITHUB_TOKEN on the host) npm run aidlc:init-check npm run aidlc:run -- --max-iterations 5
Single commands can also be run from the host: `npm run sandbox -- exec npm run aidlc:run -- --max-iterations 5`.
Results: `.ai-dlc/outputs/reports/` (final gate, unresolved issues, QCD), `.ai-dlc/outputs/reverse-docs/`, `.ai-dlc/outputs/telemetry/run-history.json`.
> Read [SECURITY.md](SECURITY.md) before a real run: the agent runs with tool permissions. Every `aidlc:*` > command refuses to run outside the sandbox (`sandbox.enforce: strict` in `aidlc.config.yaml`).
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 150 lines
- [English](README.md) | [日本語](README.ja.md)
- # ai-dlc-sdk-angular
- [](https://github.com/sun-flat-yamada/ai-dlc-sdk-angular/actions/workflows/ci.yml)
- [](https://github.com/sun-flat-yamada/ai-dlc-sdk-angular/actions/workflows/security.yml)
- [](https://angular.dev)
- [](https://docs.github.com/copilot/how-tos/copilot-cli)
- [](LICENSE)
- **An in-repository AI-DLC (AI-driven development life cycle) SDK for Angular GUIs of embedded device front panels.**
- GitHub Copilot CLI implements a screen from requirements, an HTML/CSS design and a REST API specification,
- writes and passes the unit tests, and keeps iterating — at most **N** times — until automated tests derived
- from the **human-written Excel subsystem test specification** pass. Afterwards it reverse-generates design
- documents along human interpretation axes and compares **QCD** (quality, cost, delivery) with the human
- development of the same requirement.
- ```mermaid
- flowchart LR
- subgraph inputs["Human inputs (.ai-dlc/inputs)"]
- R[Requirements]
- D["HTML/CSS design<br/>data-ui-id"]
- A["OpenAPI<br/>operationId"]
- B["ui-bindings.yaml<br/>UI ⇄ API ⇄ states"]
- X["Excel subsystem tests<br/>@ui-id"]
- end
- G0{{"Gate 0<br/>input consistency"}}
- subgraph loop["Self-healing loop (i = 1..N)"]
- C["GitHub Copilot CLI<br/>skills 01-04"]
- G1{{"Gate 1 boundary"}}
- G2{{"Gate 2 build + lint"}}
- G3{{"Gate 3 unit + coverage"}}
- G4{{"Gate 4 Excel-derived<br/>Playwright = final gate"}}
- end
- P["Reverse design docs (skill 05)<br/>QCD report (skill 06)"]
- inputs --> G0 --> C --> G1 --> G2 --> G3 --> G4
- G4 -- fail: structured failures --> C
- G4 -- pass or i = N --> P
- ```
- ## Why
- | Problem when an AI develops embedded GUIs | What this SDK does |
- | --- | --- |
- | HTML/CSS and OpenAPI do not say *which button calls which API* or *what the screen does on 503* — the AI guesses | **`ui-bindings.yaml`** binds design elements (`data-ui-id`) to API operations (`operationId`), request / response mapping, screen states and error handling; **Gate 0** checks it against the HTML, OpenAPI and Excel before any AI work |
- | The AI rewrites code it should only reuse | **Boundary policy** with five zones (`immutable`, `extensible` = append-only, `rebuildable`, `deletable`, `ignored`) enforced with `git diff` and automatic revert |
- | "Tests pass" is decided by the AI itself | The **final gate is the human Excel specification**: every case must have a passing Playwright test, expected values are hashed and protected, skipped tests fail the gate |
- | The agent runs arbitrary commands with your credentials | Every operation runs in the bundled **sandbox** (dev container): non-root, no-new-privileges, default-deny egress firewall, no host credentials. The harness **refuses to run outside it** |
- | Unbounded agent loops | **Max N iterations** with structured failure feedback, then an unresolved-issues summary |
- | AI output is hard to review | **Reverse design docs** on four human axes (reuse/delta architecture, screen & state spec, API sequences, requirement ↔ test traceability) |
- | "Is it worth it?" | **QCD report**: pass-rate trend, coverage, defects, AI credits / tokens, human effort and lead time vs. the human baseline |
- ## Quick start
- Everything runs in the bundled **sandbox** ([docs/sandbox.md](docs/sandbox.md)). The host needs only
- Docker and Node.js — no `npm ci` on the host. In VS Code, **Reopen in Container** does the same.
- ```bash
- npm run sandbox -- up # build + start the sandbox (firewall, npm ci, Copilot CLI, Chromium)
- npm run sandbox -- shell # a shell inside the sandbox; the commands below run there
- npm run aidlc:parse-excel # Excel -> .ai-dlc/outputs/generated-tests/subsystem-cases.json
- npm run aidlc:validate-inputs # Gate 0 + binding matrix (.ai-dlc/outputs/reports/binding-matrix.md)
- npm run aidlc:init-check -- --dry-run
- npm run aidlc:run -- --max-iterations 3 --dry-run # whole pipeline without calling Copilot
- ```
- Real run (needs a [GitHub Copilot](https://docs.github.com/copilot/how-tos/copilot-cli) subscription and a
- clean git tree). Copilot CLI is already installed in the sandbox:
- ```bash
- copilot login # once, inside the sandbox (or COPILOT_GITHUB_TOKEN on the host)
- npm run aidlc:init-check
- npm run aidlc:run -- --max-iterations 5
- ```
- Single commands can also be run from the host: `npm run sandbox -- exec npm run aidlc:run -- --max-iterations 5`.
- Results: `.ai-dlc/outputs/reports/` (final gate, unresolved issues, QCD), `.ai-dlc/outputs/reverse-docs/`,
- `.ai-dlc/outputs/telemetry/run-history.json`.
- > Read [SECURITY.md](SECURITY.md) before a real run: the agent runs with tool permissions. Every `aidlc:*`
- > command refuses to run outside the sandbox (`sandbox.enforce: strict` in `aidlc.config.yaml`).
- ## Use it in your existing Angular repository
- ```bash
- npm run aidlc:install -- --target ../your-panel-app --dry-run # preview
- npm run aidlc:install -- --target ../your-panel-app
- ```
- The installer copies the harness, skills, input skeleton and the sandbox (`.devcontainer/`,
- `scripts/sandbox.ts`), adds the `aidlc:*` / `sandbox` scripts and devDependencies, and never touches `src/`.
- Then run `npm run sandbox -- up` in the target. Then adapt `boundary-policy.yaml` and `.ai-dlc/references/` to your code base.
- See [docs/adoption-guide.md](docs/adoption-guide.md).
- ## Commands
- | Command | Purpose |
- | --- | --- |
- | `npm run sandbox -- <up\|exec …\|shell\|status\|down\|rebuild>` | Start / use / remove the sandbox from the host (Docker + Node.js only) |
- | `npm run aidlc:sandbox-check` | Is this process inside the sandbox (container, egress firewall, no host credentials)? |
- | `npm run aidlc:init-check` | Environment check (Node, git, Copilot CLI, Angular CLI, Playwright, inputs, skills) |
- | `npm run aidlc:parse-excel` | Excel subsystem tests → structured JSON + checklist |
- | `npm run aidlc:validate-inputs` | Gate 0: `ui-bindings.yaml` ⇄ HTML ⇄ OpenAPI ⇄ requirements ⇄ Excel; writes the binding matrix |
- | `npm run aidlc:guard` | Boundary check against the baseline (`--revert`, `--snapshot`, `--baseline <ref>`) |
- | `npm run aidlc:gate` | Gates 0-4 (`--only build,unit`, `--json`) |
- | `npm run aidlc:run` | Full loop (`--max-iterations N`, `--dry-run`, `--skip-post`, `--allow-dirty`, `--no-revert`) |
- | `npm run aidlc:reverse-doc` | Reverse design documents only |
- | `npm run aidlc:qcd` | QCD comparison report (`--run-id`) |
- | `npm run aidlc:schemas` | Regenerate the JSON Schema of `ui-bindings.yaml` |
- | `npm run aidlc:install` | Install the SDK into another Angular repository |
- | `npm run gate` | Repository quality gate (sandbox check, format, typecheck, lint, unit, harness tests, secret scan, build) |
- ## Repository layout
- ```text
- .devcontainer/ sandbox: Dockerfile, devcontainer.json, egress firewall, allow-list
- .ai-dlc/
- config/ aidlc.config.yaml, boundary-policy.yaml, schemas/
- inputs/ requirements/ ui-design/ api-specs/ ui-bindings/ subsystem-tests/ human-baseline/ (sample: REQ-NET)
- references/ existing-architecture, reusable-catalog, panel-hardware-constraints, coding-and-test-conventions
- tools/ orchestrator, copilot-cli-adapter, boundary-guard, excel-to-test-ast, validate-inputs,
- gate-runner, qcd-calculator, subsystem-kit, install, tests/
- outputs/ generated-tests/ reverse-docs/ telemetry/ reports/ (generated, git-ignored)
- .github/skills/ 01-context-and-boundary … 06-qcd-benchmark (Agent Skills for Copilot CLI)
- src/ sample "existing" Angular 22 front-panel app (800×480, zoneless, signals)
- scripts/ sandbox.ts (host launcher of the sandbox), scan-secrets.ts
- docs/ sandbox, architecture, input format, adoption guide
- ```
- ## Documentation
- - [Sandbox](docs/sandbox.md) — the isolated environment every command runs in, and how it is enforced
- - [Architecture](docs/architecture.md) — loop, gates, telemetry, safety model
- - [Input format](docs/input-format.md) — why and how to bind HTML/CSS to the REST API (`ui-bindings.yaml`, Excel notation)
- - [Adoption guide](docs/adoption-guide.md) — install into an existing repository, tune the boundary policy, run a pilot
- - [Original specification](specs/2026-11-18_scaffolding-ai-dlc-sdk-angular/spec.md) (Japanese)
- ## Status
- `0.1.0` — the harness, skills, input validation and dry-run pipeline are tested in CI. Gates 0-4 were also
- verified end to end against a reference implementation of the sample screen (not committed, because the
- sample code base must stay in its "before" state): build, lint, 37 unit tests at 99 % line coverage and 8/8
- Excel-derived Playwright cases passed, and wrong texts, skipped tests and append-only violations were
- rejected. Real Copilot CLI runs depend on your Copilot plan and environment; please report results and issues.
- ## License
- [MIT](LICENSE)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.