feat: add movie research workflow manifest

This commit is contained in:
budelius committed 2026-10-13 11:12:56 +02:00
1 parent b6de16f482
commit ab49d8d110
4 files changed
+178 -1

No files matched your search

+20
View File
@@ -0,0 +1,20 @@
name: Validate workflow manifest
on:
push:
pull_request:
workflow_dispatch:
permissions:
contents: read
jobs:
validate:
runs-on: ubuntu-latest
timeout-minutes: 5
steps:
- uses: https://github.com/actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
with:
persist-credentials: false
- name: Validate manifest
run: python3 scripts/validate_manifest.py
+51 -1
View File
@@ -1,3 +1,53 @@
# stoz3n-workflow-movies
Example repository-backed movie research workflow for stoz3n
An example **repository-backed workflow** for
[`stoz3n-chat-agent`](https://git-dev.obachan.dev/stozn/stoz3n-chat-agent).
It is declarative: the repository supplies a reviewed manifest and prompt while
the chat agent supplies the LLM, Matrix/API/MCP routing, and a small set of
host-controlled research capabilities. Repository workflows cannot execute
arbitrary code inside the bot.
## What it does
Use `/movie`, `/movies`, `@movie`, or a natural-language request routed to the
`movies` workflow. The workflow researches a film and returns:
- a spoiler-light summary;
- director(s);
- principal actors and character names when reliable;
- current streaming, rental, and purchase options for the requested country;
- clickable IMDb/TMDB, availability, and provider links.
Movie identity and credits are checked against sources such as IMDb and TMDB.
Streaming availability is treated as region- and date-sensitive and checked
against JustWatch, Watchmode, Reelgood, and official provider pages.
Example:
```text
/movie Where can I stream Arrival (2016) in Austria? Include the director and main cast.
```
## Repository contract
[`stoz3n-workflow.json`](./stoz3n-workflow.json) is the complete workflow.
The host validates and pins it by Git commit and canonical SHA-256 digest in its
`src/workflows/_repository/repository_workflows.lock.json` file. Updating this
repository does not silently change a running bot: the host lock must be
explicitly refreshed and reviewed.
Supported capabilities are intentionally narrow:
- `web_search` — Tavily search restricted to the manifest's domain allow-list;
- `web_extract` — page extraction restricted to URLs on the same allow-list.
The host owns provider credentials. No API key belongs in this repository.
## Validate
```bash
python scripts/validate_manifest.py
```
The validator uses only Python's standard library and prints the canonical
manifest digest consumed by the host lock.
+64
View File
@@ -0,0 +1,64 @@
#!/usr/bin/env python3
"""Validate this repository's stoz3n workflow manifest without dependencies."""
from __future__ import annotations
import hashlib
import json
import re
from pathlib import Path
from typing import Any
ROOT = Path(__file__).resolve().parents[1]
MANIFEST_PATH = ROOT / "stoz3n-workflow.json"
NAME_RE = re.compile(r"^[a-z0-9][a-z0-9-]*$")
DOMAIN_RE = re.compile(r"^(?:[a-z0-9](?:[a-z0-9-]{0,61}[a-z0-9])?\.)+[a-z]{2,63}$")
CAPABILITIES = {"web_search", "web_extract"}
def require(condition: bool, message: str) -> None:
if not condition:
raise SystemExit(f"manifest invalid: {message}")
def string_list(data: dict[str, Any], key: str) -> list[str]:
value = data.get(key)
require(isinstance(value, list) and value, f"{key} must be a non-empty list")
require(all(isinstance(item, str) and item for item in value), f"{key} must contain strings")
return value
def main() -> None:
data = json.loads(MANIFEST_PATH.read_text(encoding="utf-8"))
require(isinstance(data, dict), "root must be an object")
require(data.get("schema_version") == 1, "schema_version must be 1")
require(isinstance(data.get("name"), str) and NAME_RE.fullmatch(data["name"]), "invalid name")
require(bool(str(data.get("routing_description", "")).strip()), "routing_description is required")
require(bool(str(data.get("system_prompt", "")).strip()), "system_prompt is required")
for key in ("slash_commands", "mention_handles"):
values = string_list(data, key)
require(len(values) == len(set(values)), f"{key} contains duplicates")
require(all(NAME_RE.fullmatch(value) for value in values), f"{key} contains an invalid name")
capabilities = set(string_list(data, "capabilities"))
require(capabilities <= CAPABILITIES, f"unsupported capabilities: {sorted(capabilities - CAPABILITIES)}")
domains = string_list(data, "allowed_domains")
require(len(domains) <= 30, "allowed_domains exceeds 30 entries")
require(all(DOMAIN_RE.fullmatch(domain) for domain in domains), "allowed_domains contains an invalid host")
require(1 <= int(data.get("max_search_results", 0)) <= 10, "max_search_results must be 1..10")
require(1 <= int(data.get("max_tool_rounds", 0)) <= 8, "max_tool_rounds must be 1..8")
require(
1 <= int(data.get("max_concurrent_invocations", 0)) <= 20,
"max_concurrent_invocations must be 1..20",
)
canonical = json.dumps(data, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
digest = hashlib.sha256(canonical.encode()).hexdigest()
print(f"valid workflow={data['name']} sha256={digest}")
if __name__ == "__main__":
main()
+43
View File
@@ -0,0 +1,43 @@
{
"schema_version": 1,
"name": "movies",
"display_name": "Movie Research",
"slash_commands": [
"movie",
"movies"
],
"mention_handles": [
"movie",
"movies"
],
"routing_description": "Research a movie or film: identify the correct title, summarize it, list directors and principal actors, and find current region-specific streaming, rental, or purchase options with direct source links.",
"system_prompt": "You are a careful movie-research agent. Research the requested film using live web sources before answering. First disambiguate remakes and same-title films by year, director, cast, or other context. Use IMDb and TMDB for canonical identity, credits, release year, and synopsis; corroborate important facts with another reputable source. Return a concise spoiler-light summary unless the user asks for a full plot. List the director or directors and the principal cast with character names when reliable. Streaming availability is time-sensitive and country-specific: use current JustWatch, Watchmode, Reelgood, and official provider pages; state the country/region and the date checked; distinguish subscription, free/ad-supported, rent, and buy. If the user gives no region, ask for it when availability cannot be established confidently, or clearly label the region you found instead of guessing. Every availability claim must include a clickable source or provider URL. Never invent a movie, credit, provider, availability claim, or URL. If sources disagree, say so. Format the result with headings for Summary, Director, Principal cast, Where to stream, and Sources.",
"capabilities": [
"web_search",
"web_extract"
],
"allowed_domains": [
"imdb.com",
"themoviedb.org",
"justwatch.com",
"watchmode.com",
"reelgood.com",
"rottentomatoes.com",
"letterboxd.com",
"wikipedia.org",
"netflix.com",
"primevideo.com",
"amazon.com",
"disneyplus.com",
"max.com",
"hulu.com",
"apple.com",
"paramountplus.com",
"peacocktv.com",
"mubi.com"
],
"max_search_results": 8,
"max_tool_rounds": 6,
"max_concurrent_invocations": 3,
"router_selectable": true
}