feat(slack-agent): tune prompt for concise, Slack-native replies - #280
feat(slack-agent): tune prompt for concise, Slack-native replies#280JeremyFunk wants to merge 2 commits into
Conversation
The bot was writing paragraphs on paragraphs, which reads badly in Slack. Tighten the prompt's tone/length levers without touching capabilities, tool guidance, or the linking rules. - instructions.md: give "Response style" an explicit length budget (2-5 short sentences, or a top line plus 3-5 bullets), ban preamble/recap/next-step menus, require Slack mrkdwn (no `#` headers, no tables - Slack renders neither), prefer linking into Maple over pasting data, and add bad-vs-good micro-examples so the model has a shape to copy. - incident-investigation: replace the mandated six-section report (Summary / Suspected cause / Scope / Evidence / Actions / Confidence) with a ~6-line Slack shape plus an offer to expand; the long form is now opt-in when the engineer asks for a write-up. Investigation depth and evidence rules are unchanged. - dashboard-builder: cap the per-widget commentary at a sentence or two and stop it restating widget config in prose. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Your Pullfrog Router balance is exhausted. You have a payment method on file but auto-reload is disabled, so runs paused once your balance went past the overdraft buffer. Top up balance → · Enable auto-reload →
|
🍁 Maple PR previewThe preview is ready:
Deployed from |
The agent's reply string is posted through eve's Slack binding, which sets `markdownText` -> Slack's `chat.postMessage` `markdown_text` field. That field takes standard markdown, not Slack mrkdwn, so `*text*` renders as italic. Every emphasised metric would have come out italic instead of bold. Corrects the formatting guidance and the examples this PR added to use `**bold**`, and notes why inline so it isn't "corrected" back. Slack angle-bracket link syntax is left alone (pre-existing, passes through). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Your Pullfrog Router balance is exhausted. You have a payment method on file but auto-reload is disabled, so runs paused once your balance went past the overdraft buffer. Top up balance → · Enable auto-reload →
|
1 participant
The Slack bot writes "paragraphs on paragraphs" — headers, preamble, a recap of
its own investigation, and an unasked-for next-steps list. That reads badly in a
chat surface. This tunes the prompt's tone and length only; capabilities, tool
guidance, approval rules, and the "Linking into Maple" section are untouched.
Levers used
agent/instructions.md— "Response style"top-line sentence plus 3–5 bullets, with long-form gated behind an explicit
request ("write it up", "full report", "deep dive").
closing summary.
expand, rather than pre-emptively dumping everything.
model to use markdown tables and headers, neither of which Slack renders.
Now
*bold*for key numbers, backticks for ids/service names,-bullets.copy a demonstrated shape far more reliably than they follow adjectives.
agent/skills/incident-investigation/SKILL.md— this was the main source ofthe long-form output: it mandated a six-section report (Summary,
Suspected cause, Affected scope, Evidence, Suggested actions,
Confidence). Replaced with a ~6-line Slack shape — one line on what broke,
2–4 evidence bullets, one line on the first action — plus an instruction to hold
the timeline / ruled-out hypotheses / secondary evidence and offer to expand.
The section headers are now explicitly opt-in. Investigation depth, tool
selection, the evidence and no-invented-identifiers rules, and the linking
requirements are unchanged — it investigates just as hard, it just reports
shorter.
agent/skills/dashboard-builder/SKILL.md— already mostly concise; cappedper-widget commentary at a sentence or two and stopped it restating widget
config in prose (the proposal card already shows it).
Testing
apps/slack-agentparity.test.ts5/5 pass — it asserts the prompt files keepthe
maple__tool prefix and the not-linked fallback guidance, all preserved.apps/slack-agentsuite: 165 pass, 2 fail. Both failures are pre-existingCannot find module 'eve/tools'/'eve/channels/slack'resolution errors,unrelated to these markdown changes.
bun typecheckneeded.🤖 Generated with Claude Code
Need help on this PR? Tag
@codesmith-botwith what you need. Autofix is disabled.Follow-up fix (98ff18c): the formatting guidance originally said "Slack mrkdwn, not markdown"
with
*bold*. That's wrong for this transport — eve's Slack binding turns the reply string into{markdown: text}→markdownText→ Slack'schat.postMessagemarkdown_textfield, whichtakes standard markdown (and eve never calls its
gfmToSlackMrkdwnconverter on the outbound path,while the inbound path runs
slackMrkdwnToGfm— the design is symmetric). Undermarkdown_text*text*renders as italic, so every emphasised metric would have come out italic. Guidance andexamples now use
**bold**; the<url|text>link syntax is unchanged.