git:20260618.e5ad31a to git:20261013.765dea8

16 added, 15 removed. Audit A to A.

# Culture — invariable, read first

@~/master/CULTURE.md

The rules in `master/CULTURE.md` are the machine-wide source of truth and override anything below unless the below is more restrictive. The Obsidian vault at `~/master/` is this machine's central knowledge base — when this file or the project's other docs don't answer a question, query the vault first (`obsidian search` / `obsidian read` locally; `rg` / `cat` on `~/master/` in CI or headless contexts) before searching the broader internet or asking the user.

---

# CLAUDE.md — saas-api-skills

- > **Public** repo. Curated catalog of 368+ API integration skills following the [agentskills.io](https://agentskills.io/specification) spec. Sub-repo of the LastAPI workspace; root router: `../CLAUDE.md`. README owns the marketing-side narrative — this file owns the engineering / contributor view.

+ > **Public** repo. Curated catalog of 368 API integration skills following the [agentskills.io](https://agentskills.io/specification) spec. Sub-repo of the LastAPI workspace; root router: `../CLAUDE.md`. README owns the marketing-side narrative — this file owns the engineering / contributor view.

## What This Is

A community-oriented index of `*.skill.md` files, each describing how an AI agent (Claude, GPT, generic LLM) should call a specific SaaS HTTP API: endpoints, parameters, auth, example payloads. Used by:

- - `lastapi-framework` `/assistant/*` routes — every customer Lambda's auto-generated Skill Loader pulls from this catalog at deploy time

- - `lastapi-frontend` integrations browser — surfaces metadata for the 387 integrations the dashboard offers

- External agents — anyone running an MCP-style agent can clone this repo and load the skill files

+ - Public showcase of the integrations LastAPI-generated APIs ship with

- Single source of truth: the `skills/` directory. Everything else (README badges, license, contributor docs) is metadata around that.

+ **This repo is a published mirror, not the runtime source.** The canonical skill files live in `lastapi-framework` (`apigenerator/resources/1 - Project/1 - BaseProject/Project/skills/`); that copy is committed into every generated customer API and read by its `SkillLoader` (`/assistant/*` routes). Nothing fetches this repo at deploy time. The two copies share filenames but have drifted (the framework copy is newer); syncing framework → this repo is manual and currently pending (no automation exists).

## Stack

- **No runtime** — pure markdown collection, no build step, no package.json

- **Spec:** [agentskills.io](https://agentskills.io/specification) — each `*.skill.md` follows the documented frontmatter + section schema

- **Hosting:** GitHub `lastapicloud/saas-api-skills` (public, MIT licensed)

## Architecture

```

saas-api-skills

├── skills/ — 368 *.skill.md files (one per service)

│ ├── stripe-payments.skill.md

│ ├── slack-messaging.skill.md

│ ├── notion-pages.skill.md

│ └── ... 365 more

├── README.md — public-facing catalog overview + quick start

├── CONTRIBUTING.md — how to add a new skill

├── CODE_OF_CONDUCT.md — community contract

└── LICENSE — MIT

```

- Each `*.skill.md` is self-contained and consumable by an LLM with no external lookups. Categories surface via filename prefix and frontmatter `category` field (Payments, Communication, Productivity, Observability, AI & LLM, etc.).

+ Each `*.skill.md` is self-contained and consumable by an LLM with no external lookups. There is no `category` frontmatter field; README groups skills into categories by hand.

+ Frontmatter keys in use: `name`, `description`, `version`, `skill_type`, `base_url_env`, `auth_env_var`, `auth_type`, `triggers`, `license`, `metadata` (`author`, `version`, optional `documentation`), `compatibility` (`Requires <base_url_env> and <auth_env_var> environment variables.`), plus `auth_user_env` for two-credential auth. Body sections: `API Endpoints`, `Actions`, `Fields`, `Example Request Bodies` (a few files use `Example Requests`).

## Capabilities

- - **368+ service integrations** ready for AI-agent consumption (Stripe, Slack, Notion, Datadog, OpenAI, …)

+ - **368 service integrations** ready for AI-agent consumption (Stripe, Slack, Notion, Datadog, OpenAI, …)

- **Spec-compliant** — follows agentskills.io specification, so any compliant agent runtime can load these without per-skill custom code

- **Versioned via git** — every skill file's history is the audit trail

- **Public + community-extensible** — accepts PRs from external contributors per CONTRIBUTING.md

## Public Surface

| Surface | Where |

|---|---|

| Catalog browse | [`skills/`](skills/) directory + README's example table |

| Direct skill URLs | `https://github.com/lastapicloud/saas-api-skills/blob/main/skills/<name>.skill.md` (raw via `raw.githubusercontent.com/...`) |

- | Pulled by | `lastapi-framework` Skill Loader at customer-Lambda deploy time |

+ | Canonical source | `lastapi-framework` template `skills/` (this repo mirrors it) |

No HTTP API of its own — consumers fetch raw files from GitHub.

## Scopes

- **Public — open source under MIT.** External contributors welcome.

- **Catalog scope:** SaaS HTTP APIs that an AI agent might call. Not for SDK-style libraries, not for non-HTTP protocols.

- **Spec-bound:** any file in `skills/` MUST follow agentskills.io. Use CONTRIBUTING.md as the gate.

- - **Customer-Lambda contract:** changes here ripple into every deployed customer API at next deploy time. Treat schema-incompatible changes as breaking and version them deliberately.

+ - **Mirror contract:** skill content changes should land in `lastapi-framework` first, then be synced here; a change made only here never reaches customer APIs.

## Repo

- **GitHub:** `lastapicloud/saas-api-skills` (PUBLIC)

- **License:** MIT

- **Default branch:** `main`

- **Branch protection:** none (open contribution model)

- - **CI:** none today — skill files validated by reviewers + downstream consumers

+ - **CI:** `.github/workflows/pr.yml` (PRs to `develop`/`main`: changed-files, doc-sync, harness, runner budget, node checks, security audit, e2e lanes behind a single `gate` job) and `nightly-e2e.yml` (cron 07:00 UTC; currently disabled by GitHub for repository inactivity). Both call shared reusable workflows; there is no skill-schema validator yet — reviewers check frontmatter against CONTRIBUTING.md.

## Cross-Repo Dependencies

```

- saas-api-skills (this repo)

- ↓ consumed by

- lastapi-framework ← Skill Loader injects skills into every customer Lambda at deploy

- ↓

- customer Lambdas ← /assistant/skills exposes the loaded catalog at runtime

+ lastapi-framework template skills/ ← canonical; committed into every generated API

+ ↓ manual sync (pending)

+ saas-api-skills (this repo) ← public mirror

+ customer Lambdas ← SkillLoader reads their own bundled skills/; /assistant/skills exposes them

```

## Workflow

```bash

# Add a new skill

git checkout -b add-<service-name>-skill

# Author skills/<service>-<surface>.skill.md following agentskills.io

# README + CONTRIBUTING.md describe the schema in full

git commit -m "add(<service>): <surface> skill"

gh pr create --base main

```

## Rules

- Every skill must follow [agentskills.io](https://agentskills.io/specification) spec exactly — frontmatter + section ordering + parameter shape.

- - Filename pattern: `<service>-<primary-surface>.skill.md` (lowercase, hyphenated, e.g. `stripe-payments.skill.md`).

+ - Filename pattern: `<service>-<primary-surface>.skill.md` (lowercase, hyphenated, e.g. `stripe-payments.skill.md`); frontmatter `name` should equal the filename stem. Known exceptions inherited from the framework copy: `auth0-users` (`name: auth0-management`), `cloudflareDNS-domains` (`cloudflaredns-domains`), `klaviyo-marketing` (`klaviyo-marketing-campaigns`) — fix in the framework first, since `name` is the runtime skill id.

- Never embed credentials in examples — use placeholder tokens like `${YOUR_STRIPE_API_KEY}`.

- New category? Update README's example table + add a section if there are 5+ skills in the category.

- Don't force-push `main` — public history is the audit trail for all downstream consumers.