automation/publisher.py

# /// script
# requires-python = ">=3.11"
# dependencies = [
#   "huggingface_hub>=1.19.0,<2",
#   "requests>=2.32.0",
#   "requests-oauthlib>=2.0.0",
# ]
# ///
"""X publisher and reply reviewer for the August 2026 agent handover.

Live posting is opt-in. Without LIVE_POSTING=true the publish command selects
and prints the next post but never calls X. Raw replies and publication state
are written only to a private Hugging Face dataset repository.
"""

from __future__ import annotations

import argparse
import json
import os
import re
import sys
from collections import Counter, defaultdict
from datetime import date, datetime
from pathlib import Path
from typing import Any
from urllib.parse import urlparse
from zoneinfo import ZoneInfo

import requests
from huggingface_hub import HfApi, hf_hub_download
from requests_oauthlib import OAuth1


CONTROL_REPO_ID = os.environ.get(
    "CONTROL_REPO_ID", "burtenshaw/agent-handover-control"
)
CONTROL_REPO_TYPE = "dataset"
X_API_BASE = os.environ.get("X_API_BASE", "https://api.x.com/2")
URL_RE = re.compile(r"https://[^\s]+")
CAMPAIGN_SLUGS = {"campaign"}


def load_json(path: Path) -> dict[str, Any]:
    return json.loads(path.read_text())


def dump_json(payload: dict[str, Any]) -> str:
    return json.dumps(payload, indent=2, ensure_ascii=False, sort_keys=True) + "\n"


def bool_env(name: str, default: bool = False) -> bool:
    value = os.environ.get(name)
    if value is None:
        return default
    return value.strip().lower() in {"1", "true", "yes", "on"}


def hf_token() -> str | None:
    return os.environ.get("HF_TOKEN") or None


def hub_json(filename: str, default: dict[str, Any] | None = None) -> dict[str, Any]:
    try:
        path = hf_hub_download(
            repo_id=CONTROL_REPO_ID,
            repo_type=CONTROL_REPO_TYPE,
            filename=filename,
            token=hf_token(),
        )
    except Exception:
        if default is not None:
            return default
        raise
    return load_json(Path(path))


def upload_hub_json(filename: str, payload: dict[str, Any], message: str) -> None:
    HfApi(token=hf_token()).upload_file(
        repo_id=CONTROL_REPO_ID,
        repo_type=CONTROL_REPO_TYPE,
        path_in_repo=filename,
        path_or_fileobj=dump_json(payload).encode("utf-8"),
        commit_message=message,
    )


def load_queue(path: Path | None) -> dict[str, Any]:
    return load_json(path) if path else hub_json("queue.json")


def load_state(path: Path | None) -> dict[str, Any]:
    default = {
        "version": 1,
        "last_mention_id": None,
        "last_review_at": None,
        "feedback_scores": {},
        "published": {},
        "published_post_ids": [],
    }
    if path:
        return load_json(path)
    return hub_json("state.json", default=default)


def save_state(path: Path | None, state: dict[str, Any], message: str) -> None:
    if path:
        path.write_text(dump_json(state))
    else:
        upload_hub_json("state.json", state, message)


def campaign_today(queue: dict[str, Any], override: str | None = None) -> date:
    if override:
        return date.fromisoformat(override)
    timezone = ZoneInfo(queue["campaign"]["timezone"])
    return datetime.now(timezone).date()


def validate_queue(
    queue: dict[str, Any],
    catalogue: dict[str, Any] | None = None,
    check_urls: bool = False,
) -> None:
    campaign = queue["campaign"]
    start = date.fromisoformat(campaign["start_date"])
    end = date.fromisoformat(campaign["end_date"])
    expected_count = (end - start).days + 1
    posts = queue["posts"]
    errors: list[str] = []

    if len(posts) != expected_count:
        errors.append(f"expected {expected_count} posts, found {len(posts)}")

    ids = [post["id"] for post in posts]
    if len(ids) != len(set(ids)):
        errors.append("post ids are not unique")
    texts = [post["text"] for post in posts]
    if len(texts) != len(set(texts)):
        errors.append("post text is not unique")

    catalogue_slugs = CAMPAIGN_SLUGS
    if catalogue is not None:
        catalogue_slugs = CAMPAIGN_SLUGS | {
            item["slug"]
            for item in catalogue["experiments"]
            if item.get("publishable")
        }

    urls: set[str] = set()
    for post in posts:
        raw_length = len(post["text"])
        if raw_length > 280:
            errors.append(f"{post['id']} is {raw_length} raw characters")
        if post["status"] not in {"ready", "hold", "disabled"}:
            errors.append(f"{post['id']} has invalid status {post['status']}")
        if post["source_slug"] not in catalogue_slugs:
            errors.append(
                f"{post['id']} references missing/unpublishable {post['source_slug']}"
            )
        not_before = date.fromisoformat(post["not_before"])
        if not (start <= not_before <= end):
            errors.append(f"{post['id']} not_before is outside campaign")
        if post.get("fixed_date"):
            fixed = date.fromisoformat(post["fixed_date"])
            if not (start <= fixed <= end):
                errors.append(f"{post['id']} fixed_date is outside campaign")
        post_urls = URL_RE.findall(post["text"])
        if not post_urls:
            errors.append(f"{post['id']} has no public evidence URL")
        urls.update(post_urls)

    if check_urls:
        session = requests.Session()
        session.headers["User-Agent"] = "agent-handover-validator/1.0"
        for url in sorted(urls):
            try:
                response = session.get(url, timeout=20, allow_redirects=True)
                if response.status_code >= 400:
                    errors.append(f"URL {url} returned {response.status_code}")
            except requests.RequestException as exc:
                errors.append(f"URL {url} failed: {type(exc).__name__}")

    if errors:
        raise SystemExit("queue validation failed:\n- " + "\n- ".join(errors))
    print(
        f"queue_valid posts={len(posts)} start={start} end={end} "
        f"urls_checked={check_urls}",
        flush=True,
    )


def x_credentials_available() -> bool:
    return all(
        os.environ.get(name)
        for name in (
            "X_API_KEY",
            "X_API_SECRET",
            "X_ACCESS_TOKEN",
            "X_ACCESS_TOKEN_SECRET",
        )
    )


def x_session() -> requests.Session:
    missing = [
        name
        for name in (
            "X_API_KEY",
            "X_API_SECRET",
            "X_ACCESS_TOKEN",
            "X_ACCESS_TOKEN_SECRET",
        )
        if not os.environ.get(name)
    ]
    if missing:
        raise SystemExit("missing X OAuth secrets: " + ", ".join(missing))
    session = requests.Session()
    session.auth = OAuth1(
        os.environ["X_API_KEY"],
        os.environ["X_API_SECRET"],
        os.environ["X_ACCESS_TOKEN"],
        os.environ["X_ACCESS_TOKEN_SECRET"],
    )
    session.headers.update({"User-Agent": "agent-handover-august-2026/1.0"})
    return session


def x_request(
    session: requests.Session,
    method: str,
    path: str,
    **kwargs: Any,
) -> dict[str, Any]:
    response = session.request(method, f"{X_API_BASE}{path}", timeout=30, **kwargs)
    if response.status_code >= 400:
        excerpt = response.text[:1000].replace("\n", " ")
        raise RuntimeError(
            f"X API {method} {path} returned {response.status_code}: {excerpt}"
        )
    return response.json()


def classify_reply(text: str) -> list[str]:
    lowered = text.lower()
    labels: list[str] = []
    groups = {
        "reproduction_issue": (
            "reproduce",
            "error",
            "failed",
            "doesn't work",
            "does not work",
            "install",
            "command",
            "traceback",
        ),
        "experiment_suggestion": (
            "try ",
            "what about",
            "could you",
            "would you",
            "next experiment",
            "compare",
            "rerun",
        ),
        "method_question": (
            "why",
            "how",
            "which",
            "what ",
            "?",
        ),
        "claim_challenge": (
            "not enough",
            "leak",
            "overfit",
            "baseline",
            "significance",
            "cherry",
            "invalid",
        ),
    }
    for label, terms in groups.items():
        if any(term in lowered for term in terms):
            labels.append(label)
    return labels or ["reaction"]


def review_replies(
    queue: dict[str, Any],
    state: dict[str, Any],
    state_path: Path | None,
    date_override: str | None = None,
) -> None:
    today = campaign_today(queue, date_override)
    start = date.fromisoformat(queue["campaign"]["start_date"])
    end = date.fromisoformat(queue["campaign"]["end_date"])
    if not (start <= today <= end):
        print(f"review_skipped date={today} reason=outside_campaign", flush=True)
        return

    if not x_credentials_available():
        print("review_skipped reason=missing_x_oauth_credentials", flush=True)
        return

    session = x_session()
    me = x_request(session, "GET", "/users/me", params={"user.fields": "username"})
    user_id = me["data"]["id"]
    params: dict[str, Any] = {
        "max_results": 100,
        "tweet.fields": (
            "author_id,conversation_id,created_at,in_reply_to_user_id,"
            "public_metrics,referenced_tweets"
        ),
        "expansions": "author_id",
        "user.fields": "username,name",
    }
    if state.get("last_mention_id"):
        params["since_id"] = state["last_mention_id"]
    payload = x_request(session, "GET", f"/users/{user_id}/mentions", params=params)
    replies = payload.get("data", [])
    users = {
        user["id"]: user
        for user in payload.get("includes", {}).get("users", [])
    }

    label_counts: Counter[str] = Counter()
    feedback: defaultdict[str, float] = defaultdict(float)
    reviewed: list[dict[str, Any]] = []
    for reply in replies:
        text = reply.get("text", "")
        labels = classify_reply(text)
        label_counts.update(labels)
        metrics = reply.get("public_metrics", {})
        weight = 1.0 + min(float(metrics.get("like_count", 0)), 5.0) * 0.1
        lowered = text.lower()
        matched_slugs: set[str] = set()
        for post in queue["posts"]:
            if any(keyword.lower() in lowered for keyword in post["keywords"]):
                matched_slugs.add(post["source_slug"])
        for slug in matched_slugs:
            feedback[slug] += weight
        author = users.get(reply.get("author_id"), {})
        reviewed.append(
            {
                "id": reply.get("id"),
                "created_at": reply.get("created_at"),
                "conversation_id": reply.get("conversation_id"),
                "author_id": reply.get("author_id"),
                "author_username": author.get("username"),
                "text": text,
                "labels": labels,
                "matched_slugs": sorted(matched_slugs),
                "public_metrics": metrics,
            }
        )

    existing_scores = state.setdefault("feedback_scores", {})
    for slug, increment in feedback.items():
        existing_scores[slug] = round(float(existing_scores.get(slug, 0)) + increment, 4)
    if replies:
        state["last_mention_id"] = max(reply["id"] for reply in replies)
    now = datetime.now(ZoneInfo("UTC")).isoformat()
    state["last_review_at"] = now
    review_record = {
        "reviewed_at": now,
        "account": me["data"],
        "count": len(reviewed),
        "label_counts": dict(label_counts),
        "feedback_increments": dict(feedback),
        "replies": reviewed,
    }

    if state_path:
        output = state_path.parent / "latest_reply_review.json"
        output.write_text(dump_json(review_record))
    else:
        stamp = datetime.now(ZoneInfo("UTC")).strftime("%Y%m%dT%H%M%SZ")
        upload_hub_json(
            f"reply_reviews/{stamp}.json",
            review_record,
            f"review X replies {stamp}",
        )
    save_state(state_path, state, "update X reply review state")
    print(
        f"replies_reviewed={len(reviewed)} labels={dict(label_counts)} "
        f"feedback={dict(feedback)}",
        flush=True,
    )


def select_post(
    queue: dict[str, Any], state: dict[str, Any], today: date
) -> dict[str, Any]:
    today_text = today.isoformat()
    published_ids = set(state.get("published_post_ids", []))
    if today_text in state.get("published", {}):
        raise SystemExit(f"a post is already recorded for {today_text}")

    fixed = [
        post
        for post in queue["posts"]
        if post.get("fixed_date") == today_text
        and post["status"] == "ready"
        and post["id"] not in published_ids
    ]
    if fixed:
        if len(fixed) != 1:
            raise SystemExit(f"multiple fixed posts found for {today_text}")
        return fixed[0]

    eligible = [
        post
        for post in queue["posts"]
        if post["status"] == "ready"
        and post["id"] not in published_ids
        and not post.get("fixed_date")
        and date.fromisoformat(post["not_before"]) <= today
    ]
    if not eligible:
        raise SystemExit(f"no eligible post for {today_text}")
    feedback_scores = state.get("feedback_scores", {})
    eligible.sort(
        key=lambda post: (
            -float(post.get("manual_priority", 0)),
            -float(feedback_scores.get(post["source_slug"], 0)),
            int(post["sequence"]),
        )
    )
    return eligible[0]


def publish(
    queue: dict[str, Any],
    state: dict[str, Any],
    state_path: Path | None,
    date_override: str | None,
    force_dry_run: bool,
) -> None:
    today = campaign_today(queue, date_override)
    start = date.fromisoformat(queue["campaign"]["start_date"])
    end = date.fromisoformat(queue["campaign"]["end_date"])
    if not (start <= today <= end):
        print(f"publish_skipped date={today} reason=outside_campaign", flush=True)
        return

    post = select_post(queue, state, today)
    live = bool_env("LIVE_POSTING") and not force_dry_run
    if not live:
        print(
            dump_json(
                {
                    "mode": "dry_run",
                    "date": today.isoformat(),
                    "selected": post,
                    "reason": "LIVE_POSTING is not true" if not force_dry_run else "--dry-run",
                }
            ),
            end="",
        )
        return
    if not x_credentials_available():
        raise SystemExit("LIVE_POSTING=true but X OAuth credentials are missing")

    payload: dict[str, Any] = {"text": post["text"]}
    if bool_env("X_MADE_WITH_AI", default=True):
        payload["made_with_ai"] = True
    response = x_request(x_session(), "POST", "/tweets", json=payload)
    post_id = response["data"]["id"]
    username = os.environ.get("X_USERNAME", "i")
    post_url = f"https://x.com/{username}/status/{post_id}"
    record = {
        "date": today.isoformat(),
        "published_at": datetime.now(ZoneInfo("UTC")).isoformat(),
        "post_id": post_id,
        "post_url": post_url,
        "queue_id": post["id"],
        "source_slug": post["source_slug"],
        "text": post["text"],
        "x_response": response,
    }
    state.setdefault("published", {})[today.isoformat()] = record
    state.setdefault("published_post_ids", []).append(post["id"])
    if state_path:
        (state_path.parent / f"published-{today.isoformat()}.json").write_text(
            dump_json(record)
        )
    else:
        upload_hub_json(
            f"published/{today.isoformat()}-{post['id']}.json",
            record,
            f"record X post {today.isoformat()}",
        )
    save_state(state_path, state, f"record X post state {today.isoformat()}")
    print(f"published post_id={post_id} queue_id={post['id']} url={post_url}", flush=True)


def show_status(queue: dict[str, Any], state: dict[str, Any]) -> None:
    ready = [post for post in queue["posts"] if post["status"] == "ready"]
    published = state.get("published_post_ids", [])
    print(
        dump_json(
            {
                "campaign": queue["campaign"],
                "ready": len(ready),
                "published": len(published),
                "remaining": len(ready) - len(published),
                "last_review_at": state.get("last_review_at"),
                "feedback_scores": state.get("feedback_scores", {}),
                "live_posting": bool_env("LIVE_POSTING"),
                "x_credentials_available": x_credentials_available(),
            }
        ),
        end="",
    )


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser()
    parser.add_argument("command", choices=("validate", "review", "publish", "status"))
    parser.add_argument("--queue", type=Path)
    parser.add_argument("--state", type=Path)
    parser.add_argument("--catalogue", type=Path)
    parser.add_argument("--check-urls", action="store_true")
    parser.add_argument("--date", help="ISO date override for validation/dry-run tests")
    parser.add_argument("--dry-run", action="store_true")
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    queue = load_queue(args.queue)
    state = load_state(args.state)
    catalogue = load_json(args.catalogue) if args.catalogue else None

    if args.command == "validate":
        validate_queue(queue, catalogue=catalogue, check_urls=args.check_urls)
    elif args.command == "review":
        review_replies(queue, state, args.state, args.date)
    elif args.command == "publish":
        publish(queue, state, args.state, args.date, args.dry_run)
    elif args.command == "status":
        show_status(queue, state)
    else:  # pragma: no cover
        raise AssertionError(args.command)


if __name__ == "__main__":
    try:
        main()
    except KeyboardInterrupt:
        sys.exit(130)