Skip to content

t1k:discord-message

FieldValue
Modulet1k-human-comms
Version2.8.25
Effortlow
Tools—

Keywords: bullet wall, components v2, discord card, discord formatting, discord layout, discord markdown, discord message, discord status, discord-mcp, messages_send, scannable message, status update

/t1k:discord-message
<draft-or-topic> | components-v2 | example

A Discord post is scanned, often on a phone, in a channel where five other things are happening. A plain bullet wall gets skimmed past. Structure is how the content arrives, not decoration on top of it.

IntentPath
Write or fix a status / CI / release / incident updateNeeds contract → Layout rules → references/markdown-layout.md → Send
“Make this a card” / components questionreferences/components-v2.md
Start from a template (CI, incident, approval, summary)references/templates.md
Post a visual change (old + new, or new alone)modules/t1k-human-comms/rules/visual-change-discord-notice.md
Want a concrete before/afterreferences/example-platform-status.md
Prose voice — tone, no AI tellst1k-human-writing (still applies here)

Every message carries exactly one of these two literals, directly under the title/subtext. discord-message-guard.cjs (PreToolUse, shipped with the optional t1k-human-comms module, so absent where that module is not installed) reads the same strings, and every numeric limit below, from hooks/lib/discord-message-markers.json (the single source) and denies a send that breaks the contract.

What the guard lints is a data table, not prose: hooks/lib/discord-text-paths.json maps each tool to the paths holding its text. Linted (message-posting tools): messages_send, messages_edit, webhooks_execute, webhooks_edit_message, interactions_create_response (data.*), interactions_create_followup, interactions_edit_original_response, interactions_edit_followup, channels_forum_create_thread (message.*), components_v2_send, components_v2_edit, and mcp_pipeline steps whose tool is any of those. Text read: content, embed title/description/footer/author/fields, and every content string in a components tree. Primary text is content when present, else the embed/component texts joined; the marker contract applies to it, em-dashes are denied in all of it, the length cap applies to content only.

payload_json: on the five tools that take it (webhooks_execute, webhooks_edit_message, interactions_create_followup, interactions_edit_original_response, interactions_edit_followup) the string is JSON-parsed and its body is linted as its own post (the marker in content does not cover it). A payload_json that is not a JSON object is denied. Templating: mcp_pipeline {{step.path}} values are resolved by the server after the guard runs, so templated text is linted literally and warned as “not linted”.

Reminder only: components_v2_send_from_template (template text is server-side, vars.* are unbounded inputs), a poll-only post (a poll cannot carry a Needs block), any call with no extractable text. An interaction response of type 4, 5 or 7 that carries data.content IS linted; other types (modal, autocomplete, deferred update) are not message posts. Not linted, by design: tools that set names, topics, descriptions, nicknames, commands, automod, events, stage or guild settings (not message posts); components_v2_build_* and discord_intent_plan / intelligence_* (they do not send); guild_blueprint_apply (it does post a Components V2 message, but it is plan-then-apply with a dry-run default, so nothing is checked at apply time); messages_create_thread (thread name only). Button labels, select option label/description and placeholders are checked for the em-dash only (never as primary text, so they demand no marker); embed/component text over maxEmbedChars warns. Other known gaps (text below the walk’s maxDepth, which warns; tools the server adds later): references/guard-coverage.md. The rule still applies to anything unlinted; nothing checks it.

The content cap counts code points (the server’s zod .max() does too); Discord documents no unit of its own, so the cap is a contract limit, not a Discord guarantee.

WhenLiteral
Something is needed from the reader**🙋 Needs from you:** <owner> · <action> · by <deadline> — one line per ask, all three parts non-empty (maxNeedsLines in the JSON)
Nothing is needed-# ✅ No action needed (the whole line, exactly)

More than maxNeedsLines asks: one pointer line at the top (**🙋 Needs from you:** 4 asks, list at the bottom) and a numbered list last. A marker inside a code span or fence does not count and does not satisfy the contract (the send is denied). Both markers in one message is denied, as is an unterminated code fence.

  1. Title ## <subject>, then a -# subtext with scope plus the headline number (-# `main` · 16 platforms · **#46: 17/17 green**), then the Needs/No-action marker.
  2. Group by STATUS, one ### <icon> <status> heading each; drop empty groups.
  3. Icons are a legend: ✅ done/auto · 📦 manual · 🔗 link · 🔨 partial · ⚠️ blocker · 🙋 needs the reader. No decorative emoji.
  4. One line per group: **bold** names joined by ·, a code qualifier after a name.
  5. Caveats in their own -# line under the group they qualify.
  6. Actions last, numbered; ids, env vars, paths in code.
  7. Under maxChars (denied above it). Fix formatting by editing in place, never reposting.
  8. No em-dashes outside code (denied), no filler (t1k-human-writing). Bullet walls (more than maxBulletRun consecutive - lines), a missing ## title, and a post to one of the project’s report threads (discord.threads.status/ui/art, firebase.threadId) with no <@...> owner mention only warn.
  9. A project status / visual / firebase post always tags the owner <@discord.tagUserId> from .claude/t1k-project-links.json, on the Needs or subtext line, on every post, not only the final one (rules/project-status-thread.md). The guard warns (never denies) when a post to a report thread carries no <@ mention at all; it cannot know the id, so it does not check which one.

Copy-ready starting points: references/templates.md.

Do not add a sender line. discord-mcp appends -# _Sent by @<user> via discord-mcp_ itself in OAuth2 mode on messages_send, messages_publish, webhooks_execute and channels_forum_create_thread, and a second line would repeat it. bun .claude/scripts/discord-sender.cjs still resolves the session user’s id for the owner tag above.

Default to markdown through messages_send / messages_edit. It renders everywhere and can be corrected in place. messages_edit only touches this bot’s own messages; content is capped (maxChars).

Reach for Components V2 only when the message is a genuine card. Limits and the confirm gate: references/components-v2.md.

  1. Run components_v2_validate offline first.
  2. Try components_v2_send once — it is gated server-side and may return “Payload Confirmation Required”.
  3. On that gate, fall back to markdown and say why: the server runs with MCP_DRY_RUN=true, which is an operator config change, never something the agent flips.

A card cannot be edited into existence. IS_COMPONENTS_V2 is irreversible and mutually exclusive with content, so a plain message that should have been a card needs a new message.

Message layout and send-path choice for Discord via the discord-mcp server. Not authoring the substance being reported, not Discord bot administration, not other chat platforms.

  • references/templates.md — four copy-ready templates (CI/release, incident, approval, summary)
  • references/markdown-layout.md — the markdown that actually renders, and the full layout pattern
  • references/components-v2.md — card layout, container limits, the confirm gate
  • references/example-platform-status.md — a real before/after