yttrx-welcomebot/CLAUDE.md

# CLAUDE.md

Guidance for Claude Code when working on **yttrx-welcomebot** (the welcome bot
+ abuse auto-silence bot for yttrx.com).

## What this is / where it runs

A FastAPI webhook server. One Mastodon admin webhook delivers `account.created`,
`account.approved`, and `report.created` to `POST /webhook`; the app dispatches:

- **account.created** → classify the signup IP via ipapi.is; flagged
  (datacenter/vpn/proxy/tor/abuser) signups get more scrutiny (held welcome,
  ip_blocks registration, DM to moderator) before falling through to a normal
  welcome.
- **account.approved** → always DM the welcome (dedup-safe; this is what
  releases a held welcome once a human clears the approval queue). Also where
  a held, flagged signup's suspicious-watch baseline gets captured.
- **report.created** → classify the reported account, count distinct reporters,
  and auto-**silence** young/dormant accounts past the threshold — lowered for
  accounts with a flagged signup IP (see README).

A separate scheduled module, `app/suspicious_sweep.py` (run via
`python -m app.suspicious_sweep`, not part of the webhook server), suspends
any IP- or email-flagged signup that shows zero new posts and zero new
follows within `SUSPICIOUS_GRACE_HOURS` of going live. See README.md's
"Suspicious-signup sweep" section for the full policy.

| Where | What |
|---|---|
| Workstation | Source of truth: `~/yttrx-welcomebot/` (this repo) |
| `git.blairhaus.net/yttrx-welcomebot.git` | `origin` remote (HTTP Basic Auth, `GIT_AUTH_USER`/`GIT_AUTH_PASS` in `~/docker/.env` on admin — same shared cred as every other git.blairhaus.net repo, embedded in the remote URL). `.gitea/workflows/docker-build-push.yml` runs on every push to `main` (added 2026-07-06): builds the image and pushes `gitea.blairhaus.net/pmb/yttrx-welcomebot:latest` + `:<sha>` to the Gitea container registry. |
| `admin.yttrx.com` | Deploy target: `/root/yttrx-welcomebot/` is a **git checkout** of the above (converted from an rsync-deployed tree 2026-07-02), Docker container `yttrx-welcomebot` on `127.0.0.1:8087`, nginx site `hooks` → `hooks.yttrx.com`. `ssh admin.yttrx.com` logs in as **root**. Compose is **`docker-compose` v1.29.2** (hyphenated, not `docker compose`). `docker-compose.yml` now pulls `image: gitea.blairhaus.net/pmb/yttrx-welcomebot:latest` (switched from `build: .` 2026-07-06) — `root`'s `~/.docker/config.json` on this box holds a read-only (`read:package`) Gitea PAT for the pull; deploy is still `git pull` (picks up compose/`.env`-var changes) but the image itself now comes from the registry, not a local build. |
| `mammut` (`ssh mammut`) | The Mastodon instance. `tootctl`/Rails run inside the `live-web-1` container. Bot accounts + tokens live here. |

Full ops history is in secondbrain `yttrx-documentation/changelog.md`; the
user-memory note `project_welcomebot` has the current state.

## HARD RULES

- **Never overwrite the `.env` on `admin.yttrx.com`.** It holds the live
  `WEBHOOK_SECRET`, `BOT_ACCESS_TOKEN`, and `ABUSE_BOT_TOKEN`. It is untracked
  (gitignored) in the deploy checkout — `git pull`/`git reset` never touch it.
  Edit it in place for new vars.
- **Never commit secrets.** `.env` is git- and docker-ignored. Tokens/passwords
  must not land in this repo, the changelog, or CLAUDE.md.
- The bot **silences** (reversible), it does not suspend by default
  (`ABUSE_ACTION=silence`). It applies the action **without** `report_id` so the
  report stays open for human review.
- **`git` commands on admin.yttrx.com run as root** against a repo whose files
  are `chown`ed to `waffles` (see below) — root's global config there already
  has `safe.directory /root/yttrx-welcomebot` set to avoid the "dubious
  ownership" error; don't need to re-add it, but if the repo is ever
  reconstructed at that path from scratch, re-run
  `git config --global --add safe.directory /root/yttrx-welcomebot`.

## Deploy procedure (workstation → admin.yttrx.com)

```bash
# 1. Test locally (offline, no network)
cd ~/yttrx-welcomebot
python3 -m venv .venv && . .venv/bin/activate && pip install -q -r requirements.txt
WEBHOOK_SECRET=testsecret BOT_ACCESS_TOKEN=x python test_local.py   # -> ALL TESTS PASSED

# 2. Commit + push to origin (git.blairhaus.net)
git add -A && git commit -m "..."
git push origin main

# 3. Pull on admin.yttrx.com (.env is gitignored — untouched by pull/reset)
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && git pull origin main'

# 4. Re-normalise ownership if new files were added (chown is not tracked by git)
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && chown -R waffles:waffles . && \
  chown root:root .env; chmod 600 .env'

# 5. If new env vars were added this release, set them IN PLACE on the box, e.g.
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && \
  grep -q "^NEW_VAR=" .env || echo "NEW_VAR=value" >> .env'

# 6. Pull the freshly-CI-built image and restart (named volume welcomebot-data
#    persists the dedup db; the volume name is derived from the directory
#    basename, which is unchanged, so it survived the rsync -> git-checkout
#    conversion). Wait for the .gitea/workflows/docker-build-push.yml run to
#    finish first (push -> CI builds+pushes :latest) or this just re-pulls the
#    previous tag.
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && docker-compose pull && docker-compose up -d'
```

Rollback is now `git checkout <previous-sha> -- .` (or `git revert`) instead of
restoring `.bak` files. `.env.bak-*` snapshots from past deploys remain on the box; the
full pre-git-conversion tree (`/root/yttrx-welcomebot.pre-git-*/`) was removed
2026-07-02 once the git checkout was verified identical.

### Verify

```bash
ssh admin.yttrx.com 'curl -s localhost:8087/healthz'          # {"ok":true}
ssh admin.yttrx.com 'docker logs --tail 20 yttrx-welcomebot'  # clean startup
# confirm the running image + config:
ssh admin.yttrx.com 'docker exec yttrx-welcomebot python -c \
  "import app.main as m; print(m.ABUSE_DRY_RUN, m.ABUSE_ACTION, sorted(m.ABUSE_ALLOWLIST))"'
```

### Viewing logs

A `welcomebot-logs` CLI lives at `/root/bin/welcomebot-logs` on admin.yttrx.com
(source: `bin/welcomebot-logs` in this repo) — `/root/bin` is the homelab's
standard location for root-owned admin scripts (same convention as `bsd`'s
`peteftw-analytics` and the NAS backup scripts). A symlink at
`/usr/local/bin/welcomebot-logs` makes the bare command work over
non-interactive `ssh admin.yttrx.com welcomebot-logs`, since that doesn't
source `~/.bashrc` and so wouldn't otherwise see `/root/bin` on `PATH`. It
wraps `docker logs` for the container:

```bash
ssh admin.yttrx.com welcomebot-logs                 # last 200 lines
ssh admin.yttrx.com welcomebot-logs -f --abuse      # follow abuse activity only
ssh admin.yttrx.com welcomebot-logs -n 1000 --since 24h
ssh admin.yttrx.com welcomebot-logs --ip --since 24h  # signup-IP scrutiny decisions
ssh admin.yttrx.com 'welcomebot-logs --help'
```

### Inspecting signup-IP classification history

`welcomebot-signups` (`/root/bin/welcomebot-signups`, symlinked from
`/usr/local/bin` same as above; source `bin/welcomebot-signups`) reads the
persistent `signup_ip` sqlite table directly — a permanent record of every
classification decision, independent of log retention:

```bash
ssh admin.yttrx.com welcomebot-signups              # last 20 decisions
ssh admin.yttrx.com welcomebot-signups --flagged    # only datacenter-flagged signups
ssh admin.yttrx.com welcomebot-signups -a someuser
ssh admin.yttrx.com welcomebot-signups -n all
```

Both CLIs are just files tracked in this repo now — after editing either one,
reinstall with:

```bash
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && git pull origin main && \
  install -m755 bin/welcomebot-logs bin/welcomebot-signups /root/bin/'
# /usr/local/bin/welcomebot-{logs,signups} are symlinks to /root/bin — no
# need to touch them again once created.
```

### Rollback

```bash
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && \
  cp app/main.py.bak-YYYYMMDD app/main.py && cp .env.bak-YYYYMMDD .env && \
  docker-compose up -d --build'
```

## Rollout safety: dry-run

`ABUSE_DRY_RUN=true` makes the abuse handler log + DM what it *would* do without
silencing anyone. The shipped `.env.example` default is `true`; **production is
currently `false` (live and acting)**. To re-enter dry-run:

```bash
ssh admin.yttrx.com 'cd /root/yttrx-welcomebot && \
  sed -i "s/^ABUSE_DRY_RUN=.*/ABUSE_DRY_RUN=true/" .env && docker-compose up -d'
```

## The moderator bot token (gotcha)

The abuse handler posts to `POST /api/v1/admin/accounts/:id/action`, which needs
the **`admin:write:accounts`** OAuth scope — *separate* from `admin:write:reports`.
The bot account (`bot`) is an **Admin** (so it has the role permissions), but a
token created in the UI may omit `admin:write:accounts` → silence returns `403`.

**IP-scrutiny's `register_ip_block` needs an additional scope,
`admin:write:ip_blocks`**, to `POST /api/v1/admin/ip_blocks` — mint (or re-mint)
the token with it included, or `IP_SCRUTINY_AUTO_IPBLOCK` writes will 403 (logged
as an error; nothing else in the bot is affected).

**Email-domain scrutiny's `register_email_domain_block` needs another scope,
`admin:write:email_domain_blocks`**, to `POST /api/v1/admin/email_domain_blocks`
— same deal: mint (or re-mint) the token with it included, or
`CHECK_MAIL_AUTO_DOMAIN_BLOCK` writes (both the signup-time path and the
report-triggered suspend override) will 403 (logged as an error; the suspend
itself still goes through since that only needs `admin:write:accounts`).

**The suspicious-signup sweep (`app/suspicious_sweep.py`) needs another scope,
`admin:read:accounts`**, to `GET /api/v1/admin/accounts/:id` (fetches current
post/follow counts + role/suspended state to decide whether to suspend) —
without it every sweep run 403s per-account (logged as an error, that account
is simply retried next run; nothing else in the bot is affected).

Mint a correctly-scoped token from the Rails console on mammut:

```bash
ssh mammut 'docker exec $(docker ps -f name=live-web-1 -q) bin/rails runner "
acct = Account.find_local(%q{bot})
scopes = %q{read:statuses write:statuses admin:read:reports admin:write:accounts admin:write:ip_blocks admin:write:email_domain_blocks admin:read:accounts}
app = Doorkeeper::Application.create!(name: %q{Abuse handler vN}, scopes: scopes, redirect_uri: %q{urn:ietf:wg:oauth:2.0:oob})
tok = Doorkeeper::AccessToken.create!(application_id: app.id, resource_owner_id: acct.user.id, scopes: scopes)
puts tok.token
"'
```

Put the result in `ABUSE_BOT_TOKEN` on the box's `.env`, restart, and revoke the
old token (`Doorkeeper::AccessToken.find(<id>).revoke`).

**Harmless write-permission probe** (no real account touched — `403` = missing
scope, `404` = authorized):

```bash
curl -s -o /dev/null -w "%{http_code}\n" -X POST \
  -H "Authorization: Bearer $ABUSE_BOT_TOKEN" -d "type=none" \
  https://yttrx.com/api/v1/admin/accounts/0/action
```

Same idea for the `admin:write:ip_blocks` scope IP-scrutiny needs (`403` =
missing scope; `422` = authorized, just missing/invalid params — no block is
created either way):

```bash
curl -s -o /dev/null -w "%{http_code}\n" -X POST \
  -H "Authorization: Bearer $ABUSE_BOT_TOKEN" \
  https://yttrx.com/api/v1/admin/ip_blocks
```

And for `admin:write:email_domain_blocks` (`403` = missing scope; `422` =
authorized, just missing/invalid params):

```bash
curl -s -o /dev/null -w "%{http_code}\n" -X POST \
  -H "Authorization: Bearer $ABUSE_BOT_TOKEN" \
  https://yttrx.com/api/v1/admin/email_domain_blocks
```

And for `admin:read:accounts` (`403` = missing scope; `200`/`404` = authorized):

```bash
curl -s -o /dev/null -w "%{http_code}\n" \
  -H "Authorization: Bearer $ABUSE_BOT_TOKEN" \
  https://yttrx.com/api/v1/admin/accounts/0
```

## Mastodon side (mammut)

- The webhook (Administration → Webhooks) is already subscribed to
  `account.created`, `account.approved`, `report.created` — no change needed for
  code redeploys.
- Staff are never auto-silenced: `ABUSE_SKIP_PRIVILEGED=true` reads the payload
  `target_account.role` (skips any assigned staff role, id > 0), plus the
  explicit `ABUSE_ALLOWLIST`. Current staff: `waffles`/`tommertron` (Owner),
  `davis`/`bot` (Admin).

## Appeals page

Silenced users are DMed `ABUSE_HELP_URL` =
`https://welcome.yttrx.com/posts/account-limited/`. That page is a Hugo
(Compost) content file at `admin.yttrx.com:/var/www/html/welcome/content/posts/`,
rebuilt with `./build.sh` in that dir (see secondbrain `misc-sites.md`).