codecel-discord-bot

A Discord bot that runs bug reports for a community that ships games. Players report bugs in a forum channel, Claude checks each report, and the bot sends it to the developer who owns that game. Developers work from one board: browse, mark fixed, grab a transcript for their AI agent, and post update notes.

Built with Python (discord.py), SQLite and the Anthropic API (Claude Haiku by default, well under a cent per report).

What it does

Reports

• Checks every new post in your bug forum against your template. Missing info gets a friendly list and the Needs Info tag, and the bot re-checks when the reporter answers. Posts left unanswered for 3 days are closed.
• Puts the game in the post's title, sets the device tag, numbers the post (#12 title), spots duplicates (and bugs that were already fixed), and warns about personal info in screenshots.
• Exploit write-ups posted in the public bug forum are removed at once (testers are different: an exploit sent with /bug in the testing channel is kept and shown like any bug, so other testers can reproduce it) and appear as a high priority card on the owner's board. The reporter is told by DM and gets a DM on every status change.
• Trolling gets a strike ladder: warning, 1-day mute, kick, 7-day ban, permanent ban. Kicks and bans need a developer's confirmation. Staff are never struck.

Dev boards (one forum post per dev)

• A summary of the dev's open bugs plus two dropdowns: open a game's bugs in a private page (screenshots, videos, status buttons), or download a transcript.
• A new bug gives the dev a ghost ping that they can turn off with /pings.
• Fixed bugs: the reporter (and everyone who reacted 👀) is pinged, the post closes and is deleted after 24 hours. A reply of "still broken" reopens it.
• Transcripts: a zip per bug or per game with the conversation, screenshots, full videos and a few video frames (via ffmpeg), so an AI agent can read it. Files that do not fit Discord's upload limit become download links.
• Fix log: a pinned post with fixed / open counts per game and every fix with its transcript.
• Update announcements: after 3 fixes the bot drafts an update. Devs press Post, Edit or Later, or use /update. What a dev types is rewritten by Claude into clean patch notes.

Suggestions (optional, SUGGESTIONS_CHANNEL_ID and DEV_SUGGESTIONS_FORUM_ID): the same idea for a forum where players suggest improvements. Claude checks each post, names the game, asks for what is missing, spots duplicates and sends it to the dev who owns the game, on a suggestion board in its own dev-only forum (with the 🙏 votes). Devs type 12 added or 7 declined, or use the buttons; the player is pinged and the post is deleted after 24 hours. Added suggestions go into the update announcements next to the bug fixes. A bug posted there becomes a normal bug and the player is told. The bot creates the Added / Declined status tags in that forum itself.

Reports by command: anyone can use /bug (a short form, up to 10 files and a links box) or /suggest (the same). More files and links go in as replies in the post, with no limit. The bot makes the post in the bug or suggestions forum for the player and then handles it like a post written by hand. Testers using /bug in the testing channel get a thread there instead, and the bug goes straight to the owning dev's board.

Offline-safe: when the bot starts it catches up on everything it missed (new posts, replies, timers, deletes).

Commands

Command	Who	What
/bug	testers (in the tester channel)	report a bug with a form
/pings	devs	choose whether you are pinged for new bugs and for exploit reports
/update	game owner, admins	post an update with the pending bug fixes
/config set / show	game owner, admins	per game: link, icon, announcement channel, ping role, owner
/release (new, quiet, edit, undo)	game owner, admins	move a game between the Released and Unreleased lists (/release takes an optional play link)
/games	admins	post the Released / Unreleased games board in this channel. It updates itself
/strikes, /pardon	devs	see or remove troll strikes
/botrules	admins	post the rules and the strike ladder
/preview	admins	post sample embeds in the test channel

Permissions: devs act on the games they own (any dev on games nobody owns), jr devs are read only, admins can do everything. Roles are checked on the server for every click.

Setup

1. Create the Discord application

1. https://discord.com/developers/applications > New Application. Copy the Application ID.
2. Bot tab > Reset Token (copy it once). Turn on Message Content Intent and Server Members Intent.
3. OAuth2 > URL Generator: scopes bot and applications.commands. Easiest permission: Administrator. Otherwise grant View Channels, Send Messages (also in Threads), Create Public Threads, Read Message History, Embed Links, Attach Files, Manage Messages, Manage Threads, Manage Channels, Manage Roles, Moderate Members, Kick Members, Ban Members and Mention Everyone.
4. Invite it, then drag the bot's role above the roles it may punish.

2. Prepare the server

• A forum channel for reports, with tags: Mobile, Computer, Controller, and the status tags Investigating, Needs Info, Fix in Progress, Fixed, Duplicate, Not a Bug. Mark the status tags as moderated so only the bot sets them. Games are not tags (a forum holds 20 at most): the game is in each post's title, and /bug lists the released games.
• A forum channel only devs can see (the boards).
• A text channel for the mod log, and optionally a tester channel and an updates channel.

3. Configure

cp .env.example .env     # then fill it in; every key is explained in the file

Turn on Developer Mode in Discord to copy channel, role and server IDs.

4. Run it

python -m venv .venv
.venv/bin/pip install -r requirements.txt      # Windows: .venv\Scripts\pip
.venv/bin/python -m bot                        # Windows: .venv\Scripts\python

Needs Python 3.12+ and ffmpeg on the PATH (ffmpeg is only used for video frames in transcripts). On Windows start.cmd restarts the bot if it crashes.

Hosting on Oracle Cloud (or any Ubuntu VPS)

The bot only makes outgoing connections, so no ports or firewall rules are needed. It stores everything in data/ (bot.db and the fix archive).

Option A: systemd (simplest). On a fresh Ubuntu 22.04 / 24.04 VM (x86 or ARM Ampere):

git clone https://github.com/foodcel/codecel-discord-bot.git
cd codecel-discord-bot
bash deploy/oracle-setup.sh     # installs Python + ffmpeg + a virtualenv, creates .env, run it again after filling .env

The second run installs and starts a codecel-bot service. Then:

journalctl -u codecel-bot -f                       # logs
sudo systemctl restart codecel-bot                 # restart
git pull && .venv/bin/pip install -q -r requirements.txt && sudo systemctl restart codecel-bot   # update

Option A2: no sudo password (you can log in but cannot use sudo). Needs Python 3.12+ (with the venv module) already on the VM:

git clone https://github.com/foodcel/codecel-discord-bot.git && cd codecel-discord-bot
bash deploy/user-install.sh     # runs the bot as a systemd user service, no root needed
journalctl --user -u codecel-bot -f

Option B: Docker

cp .env.example .env && nano .env
docker compose up -d --build
docker compose logs -f

Deploying changes from your PC: copy deploy/vm.env.example to deploy/vm.env, fill it in, commit, then run bash deploy/deploy.sh. It pushes, updates the VM, restarts the bot and waits until it logs in.

The old JavaScript version: the bot used to be written in JavaScript. That version is the git tag js-final. bash deploy/rollback-js.sh puts it back on the VM (same database). To return to Python afterwards run git checkout main on the VM and then bash deploy/user-install.sh there.

Moving an existing bot (keeps your reports, strikes and archive): stop the old bot, copy .env and the whole data/ folder to the VM (scp -r data .env user@vm:codecel-discord-bot/), then start it there. Never run two copies with the same token at once, they would both answer.

Oracle notes:

• Always Free VMs can be reclaimed when they sit idle (CPU, network and memory all under about 20% for 7 days). A quiet bot looks idle, so upgrade the account to pay-as-you-go (Always Free resources stay free) or check on it now and then.
• Back up data/ regularly (stop the bot or use sqlite3 data/bot.db ".backup bot-backup.db").
• Keep .env private (chmod 600 .env). If a token ever leaks, reset it in the developer portal.

Project layout

bot/__main__.py            start-up and Discord events (python -m bot)
bot/core/                  config (env.py: every id), database, timers, catch-up, AI calls, permissions,
                           access.py (who may use which command), links.py (every link check)
bot/discord_ui/            slash command definitions and routing (interactions.py: one table per button / dropdown / form),
                           the two looks (look.py, pngs.py, ui.py), /help, the dev log
bot/features/bugs/         reports: intake (Claude), forms, tester reports, numbers, tags, dev boards, fixes, fix log, transcripts, strikes
bot/features/updates/      update drafts and announcements, guides, announcement texts
bot/features/games/        game lists, /release and its preview, the library, the Roblox watcher, the bounty board
bot/features/testers/      the tester roster and the tester board
bot/features/invites/      the invite tracker and /invites
bot/features/fun/          /mog
scripts/                   admin and diagnostic tools run by hand (lock_commands.py sets the per-role command rules)
tests/                     tests for the logic that needs no Discord (deploy.sh runs them with a lint check)
deploy/                    systemd unit, Oracle setup, deploy and rollback scripts
docs/                      DESIGN.md (the two looks), INVITES.md, AUDIT.md

Safety notes

• Report text is untrusted. Claude only returns structured data, and the bot's own code decides every action, so a report cannot talk the bot into banning someone.
• Strikes require high confidence, and kicks and bans always need a developer to confirm.
• A daily cap (AI_DAILY_CAP) limits how many Claude calls can be made.

License

MIT