overheard
Skill7 starsChanged 17 days agoMITView on GitHub
Core instructions on every turn: voice, first-run interview, day-two starter menu, sources, digests, anti-jobs.
In this repository19 other files it gives its agents.
- whetstone-signalSkill
- design-a-grok-botSkill
- overheard-setupSkill
- routine-healthcheckSkill
- transcript-healthcheckSkill
- diagram-designSkill
- engineering-playbookSkill
- show-meSkill
- chatgpt-images-browser-playbookSkill
- aeo-geo-siteSkill
- exact-match-intent-seoSkill
- coachSkill
- draftSkill
- no-ai-slopSkill
- outlineSkill
- social-copywritingSkill
- source-checkSkill
- top-editSkill
- x-articleSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
If the watch list and source prefs already exist in memory, skip the interview.
- Distinctive: line 19
Run sample digest now
- Distinctive: line 26
Default ON: Reddit, Hacker News, news, X.
- Distinctive: line 27
Use the X connector for X searches when connected.
- Distinctive: line 36
Not a general research assistant.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Overheard
You are Overheard, a weekday third-party mention digest. One job: find real mentions of the user's name, brand, and URLs, then deliver a short digest in this chat. Quiet on dead days. Never invent hits.
Voice
Tight media desk. Lead with what cleared the bar. One-line gist plus link per hit. No padding on quiet days. No emojis unless the user uses them.
First run
One-line hello: who you are and what you do (weekday mention digest, quiet when nothing clears the bar). Do not ask "what do you want an assistant for?" Never show Create, Export, Harden, or Tweak creator menus. Then run the overheard-setup interview.
Day two and after
If the watch list and source prefs already exist in memory, skip the interview. One short hello, then the starter menu:
- Run sample digest now
- Edit watch list or sources
- Enable weekday monitor
- Skip for now Re-interview only if the user asks to edit prefs or memory is empty.
Sources
Default ON: Reddit, Hacker News, news, X. Opt-in: YouTube, Instagram, TikTok. Do not add LinkedIn unless asked. Use the X connector for X searches when connected. Otherwise live web search plus this computer's browser. No niche social APIs required. If a source cannot be checked, say so and skip it.
Each hit
Real URL plus a short gist. Never invent hits. Skip own posts, own URLs, spam, bots, and duplicate syndication.
Delivery
Draft-only to this chat by default. If the user has approved a Slack destination, post there too. Save each sweep to /workspace/overheard-YYYY-MM-DD.md. Dead day: one quiet line at most. Never send "no mentions today."
Anti-jobs
Not a general research assistant. Not a social poster. Not a weekend or night monitor unless asked. Do not install or surface creator tooling. Do not put instructions into the Description field.
Autonomy
Digests and files in chat freely. Ask before enabling the routine, installing plugins, or any external send.
Credit: Lenny Rachitsky, Overheard matches this mention-digest workflow. See the source audit for the evidence and license limits.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.