メニュー
Say what you are doing and what you find, in the Slack thread you were started in. TRIGGER within the first minute of every turn, again on every action you take, and with --notify the moment you discover something.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/OpenRouterLabs/ori-feature-catalog --skill slack-statusコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Attach ANY file to the current Slack thread — text or binary. A log, a diff, a CSV, a long answer as a snippet, and equally a PDF, a PNG, a zip, a sqlite db, a tarball: the bytes go up unchanged. TRIGGER when the user asks you to attach, upload, or send a file, or when an answer is too long or too structured to read as a message (a full log, a wide table, hundreds of lines of output) and a file reads better than a wall of text. Also TRIGGER when the user asks for something "as a file", "as a snippet", "as an attachment", or "upload that". Write the content to a file first with Bash, then attach it. This is NOT for charts (use slack-chart) or generated images (use slack-image) — those render and upload on their own. This is NOT for ordinary replies: a short answer belongs in the message, not in an attachment.
Move THIS conversation onto a fresh Slack thread, keeping the same agent session and everything it remembers. ONLY TRIGGER when the user EXPLICITLY asks to continue the current conversation somewhere else — e.g. "continue this in a new thread", "move this to a fresh thread", "same conversation, new thread", "this thread is too long, carry on in a new one". NEVER trigger on your own initiative. Never carry because a thread looks long, cluttered, or busy — if the user did not ask, keep replying where you are. This is NOT for starting separate work: if the user wants a NEW task in its own thread with a fresh start, that is the spawn-thread skill instead. Carrying MUTES the thread you are in, so doing it unasked strands whoever is talking to you.
Read Slack from the agent — thread replies, channel history, user lookup and mention resolution, and opening a DM. Channel and thread resolve env-first ($SLACK_CHANNEL_ID / $SLACK_THREAD_TS) with a --channel fallback, so you can usually omit them. TRIGGER when asked to fetch a thread or channel history, find someone's user id, resolve a mention, or DM someone. This skill does NOT post, edit, delete or react — your answer is just the text you write, and progress goes through the slack-status skill.
SKILL.md を表示中
| name | slack-status |
| description | Say what you are doing and what you find, in the Slack thread you were started in. TRIGGER within the first minute of every turn, again on every action you take, and with --notify the moment you discover something. |
# the live line — free, silent, replaced by your next one. Keep it current.
bun features/slack/skills/slack-status/scripts/index.ts "reading run-events.ts to find where AgentFailure comes from"
# a finding — permanent, pings the thread, and still sets the line
bun features/slack/skills/slack-status/scripts/index.ts --notify "It is not the code: .ori/sdk is generated per machine, and yours predates AgentFailure"
The thread comes from your environment. You never pass it.
A greeting, an acknowledgement, a correction, a question you can answer off the top of your head — those are finished before any status could land. I'll say hello back followed by hello is two messages where one was wanted, and the thread keeps the useless one. If you are about to reply, reply.
Everything below is for a turn that is still running a minute from now.
--notify. A cause found, a decision made, a number that changes what happens next. That one is permanent and people can read it later; the plain line dies with your next update.A run that works in silence is indistinguishable from a crash.
Checking CI on the 9 red PRs — 6 are the same flaky timeout, looking at the other 3, not checking CI. A bare label tells nobody anything.
The line is capped at 120 characters — Slack renders it on one line and never folds it. A --notify message may run to 300. Over the cap it is rejected with the reason, not cut: shorten and send again.
Lead with the point. The line shows in full under the composer, but the copy inside the thread — the one people actually read while they wait — is shortened to about 40 characters, on a word boundary. So put the subject first: rebasing the 7 conflicting PRs survives whole, while going through the list of open pull requests to rebase the 7 that conflict reaches the reader as going through the list of open… and says nothing.