agentleFS
freskhufreskhu/genesis/.claude/skills/forum-pulse/SKILL.md

forum-pulse

Skill5 starsChanged 6 months agoApache-2.0View on GitHub

Read the upstream Genesis community (Discussions + Issues), cross-reference with the user's MemPalace, and surface what's actually relevant to *their* work. The connective tissue between the user's local Agentic OS and the swarm.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 6 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Ensure gh CLI is authenticated:

    line 16
  2. Very specific:

    If not — instruct the user: gh auth login.

    line 20
  3. Very specific:

    Read Database/genesis_config.json for last_pulse_check (timestamp of previous run).

    line 25
  4. Distinctive:

    Run a semantic search against the user's palace:

    line 62
  5. Very specific:

    Run /hire's validation step (Lena registers in team_members, Sarah reviews scope).

    line 122
  6. Distinctive:

    Don't auto-activate — just install.

    line 123
  7. Very specific:

    If the user opted into auto-sync during /genesis (see auto_sync_enabled), suggest also scheduling forum-pulse daily:

    line 166
  8. Distinctive:

    Every comment, reaction, or PR opened by the agent ends with:

    line 188
  9. Very specific:

    Kill switch: if user runs /forum-pulse --pause, halt all autonomous actions until explicit /forum-pulse --resume.

    line 214
  10. Very specific:

    Audit: every autonomous action logged to activity_history with action='forum_post' or 'forum_reaction' or 'forum_pr_draft'.

    line 215
  11. Distinctive:

    If the agent is uncertain about an action's appropriateness — even at Level 4 — drop to "propose, ask user" mode for that single item.

    line 219
  12. Distinctive:

    Don't double-count: If a thread was already proposed in a previous run, skip it unless something material changed (new comment, new label, …

    line 228
  13. Very specific:

    If they just ran /genesis and palace is sparse, the relevance check has nothing to cross-reference.

    line 236
  14. Distinctive:

    Always end with the action list (not just the digest).

    line 244

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Forum Pulse

Most online communities die because nobody has time to read everything. This skill is the user's filter — every day (or on demand), it scans the upstream community at freskhu/genesis and surfaces only what matters to them.

The signal: the user's own work, captured in MemPalace, is the lens. Generic noise gets filtered. Relevant agents, threads, bugs, or proposals get a spotlight.


Step 0 — Pre-flight

  1. Ensure gh CLI is authenticated:
    gh auth status
    If not — instruct the user: gh auth login. Skill cannot proceed without it.
  2. Confirm MemPalace is reachable:
    python3 scripts/palace.py status
  3. Read Database/genesis_config.json for last_pulse_check (timestamp of previous run). If missing, default to "7 days ago".

Step 1 — Fetch new community activity

Pull everything since last_pulse_check:

SINCE="2026-05-01T00:00:00Z"  # from config

# Discussions (use GraphQL — REST API doesn't expose discussion content well)
gh api graphql -f query='
{
  repository(owner: "freskhu", name: "genesis") {
    discussions(first: 50, orderBy: {field: UPDATED_AT, direction: DESC}) {
      nodes {
        number title bodyText updatedAt category { slug }
        comments(first: 5) { nodes { bodyText author { login } } }
      }
    }
  }
}' > /tmp/discussions.json

# Issues
gh issue list --repo freskhu/genesis --state all --limit 50 \
  --json number,title,body,labels,state,updatedAt,comments > /tmp/issues.json

Filter to items updated after last_pulse_check.


Step 2 — Cross-reference with the user's palace

For each discussion/issue title + body:

  1. Run a semantic search against the user's palace:
    python3 scripts/palace.py search "{title + first 200 chars of body}" --mode vector --limit 3
  2. Capture the top match's distance score and which wing/room it came from.
  3. Relevance bucket:
    • HIGH: semantic distance < 1.5 — the user has clearly worked on something semantically close.
    • MEDIUM: distance 1.5–2.2 — adjacent topic, might be useful.
    • LOW: distance > 2.2 — generic interest only.

Also boost relevance for:

  • Agent Marketplace category — always at least MEDIUM (agents are reusable).
  • Issues affecting the user's pinned skills (read Team/orchestrator.md for which skills the user pinned during /genesis).
  • Breaking change announcements — always HIGH.

Step 3 — Synthesise the digest

Format:

Forum pulse — 2026-05-15 · 7 days · 12 new threads

  ━━ HIGH RELEVANCE ━━
  💬 #42  Discussion · Agent Marketplace
      "Lucia v2 — better LinkedIn voice tuning"
      → matches your "product marketing" wing (semantic dist 0.87)
      → 3 comments, last from @joana 2h ago

  🐛 #29  Issue · BUG · open
      "/dream consolidates wrong tier when palace > 10k drawers"
      → you have 12,400 drawers (matches the trigger condition)
      → fix in PR #31, ready to merge

  ━━ MEDIUM RELEVANCE ━━
  💡 #38  Discussion · Ideas
      "Cross-OS agent transfer protocol"
      → adjacent to your /sync-upstream usage
      → conversational, no commitment

  ━━ LOW (showing top 2) ━━
  💬 #45  General · "What hardware do you run on?"
  💡 #44  Polls · "Default language for new installs"

  Total noise filtered: 6 threads.

Step 4 — Propose actions

For each HIGH item, propose a concrete action and ask for confirmation:

Agent share (Marketplace)

"Want me to import the agent definition from #42 into .claude/agents/? I'll review it first, sanitise, and put it under your control."

If yes:

  • Fetch the agent block from the discussion thread.
  • Write to .claude/agents/{name}.md.
  • Run /hire's validation step (Lena registers in team_members, Sarah reviews scope).
  • Don't auto-activate — just install.

Issue affecting the user

"Issue #29 directly hits a condition in your setup. Want me to: (a) subscribe you to the thread (so you see updates), (b) cherry-pick the fix from PR #31 now via /sync-upstream, (c) leave it for now?"

Discussion alignment

"You worked on Y last week. There's a discussion #38 exactly on Y. Want me to draft a comment with your perspective for you to review and post?"

If yes:

  • Draft comment in outbox/Forum Pulse/draft-comment-{discussion-number}.md.
  • Include: the user's relevant past work (citations to palace), a one-paragraph contribution.
  • User reviews and posts manually with gh discussion comment {n} (the skill never posts on the user's behalf without explicit click-through).

Step 5 — Persist

# Update config
jq '.last_pulse_check = now | tojson' Database/genesis_config.json > /tmp/cfg.json
mv /tmp/cfg.json Database/genesis_config.json

Log:

INSERT INTO activity_history (actor_id, action, entity_type, summary)
VALUES (1, 'forum_pulse', 'system',
        'Pulse run: N new threads, M high relevance, K actions proposed.');

Save the digest itself to MemPalace (so future searches can reference what was seen):

python3 scripts/palace.py add --wing community --room reference --hall hall_events \
  --content "{full digest text}" --source "forum-pulse"

Step 6 — Optional: schedule daily

If the user opted into auto-sync during /genesis (see auto_sync_enabled), suggest also scheduling forum-pulse daily:

/schedule daily 09:00 /forum-pulse

Autonomy levels

By default, the skill is read-only — every action requires user confirmation. But the user can raise the autonomy level in Database/genesis_config.json (forum_pulse_level field). Each level inherits the previous.

LevelNameWhat the agent can do without asking
0Lurker (default)Read, classify, propose actions. Never post or import anything.
1ReactorAdd emoji reactions (👍 👎 🎉 ❤️) on threads marked HIGH relevance. No text.
2DrafterWrite comment drafts to outbox/Forum Pulse/draft-{n}.md. User reviews and posts manually.
3SpeakerPost low-stakes comments directly: "+1" with one-line context on Ideas, factual answers to Q&A where the palace has clear evidence, "I see this too" on bug reports matching the user's setup. Always signed.
4ContributorOpen PR drafts for trivial fixes (typos, doc edits) referenced in Issues. Comment substantively in Discussions when the palace gives strong signal. Always signed.

Signature for any auto-post (level 3+)

Every comment, reaction, or PR opened by the agent ends with:

---
*Posted by `@{user}`'s Genesis instance ({skill_that_triggered}). Reply here or open a discussion to push back.*

Honesty matters — readers always know it's AI-assisted, never disguised as the user's direct typing.

Rate limits per level

LevelMax posts/weekReactions/weekPR drafts/month
0000
10200
20 (drafts only)200
35300
410504

Hard caps. Counts kept in genesis_config.json.forum_pulse_counters. Reset weekly/monthly.

Hard skip rules (apply at all levels)

  • Don't reply to your own threads — if user is OP, never auto-engage.
  • Don't post if user already replied in last 24h — they're on it.
  • Don't post in heated threads — if the last 3 comments contain disagreement markers (disagree, wrong, actually, downvote ratio rising), pause and surface to user instead.
  • Kill switch: if user runs /forum-pulse --pause, halt all autonomous actions until explicit /forum-pulse --resume. Persisted across sessions.
  • Audit: every autonomous action logged to activity_history with action='forum_post' or 'forum_reaction' or 'forum_pr_draft'.

When to escalate

If the agent is uncertain about an action's appropriateness — even at Level 4 — drop to "propose, ask user" mode for that single item. Better to over-ask than to over-post.


Anti-patterns

  • At Level 0 (default): never post or import anything without per-item confirmation. Drafting is fine; posting is the user's.
  • Even at Level 4: importing agents requires a confirmation click. Marketplace agents can be malicious — always show the SKILL.md content first.
  • Don't surface every thread. The whole point is filtering. If everything is "MEDIUM", drop the digest entirely and tell the user "nothing relevant this run."
  • Don't double-count. If a thread was already proposed in a previous run, skip it unless something material changed (new comment, new label, state change).
  • Respect GitHub rate limits. 5000 req/h authenticated — cache responses for 1h on repeat calls in same session.
  • Never disguise AI as human. Any auto-post at L3+ MUST carry the signature footer. No exceptions.

When the user has no palace yet

If they just ran /genesis and palace is sparse, the relevance check has nothing to cross-reference. In that case:

  • Show top 5 trending threads regardless of relevance.
  • Mention: "I'll get smarter at this once you've had the system for a few weeks — your work history is the filter."

Output format reminder

Always end with the action list (not just the digest). The digest is for context; the actions are for the user to decide.

Actions proposed:
  [1] Import agent Lucia v2 from #42
  [2] Cherry-pick fix from PR #31 (resolves #29)
  [3] Draft comment for discussion #38
  [4] Skip everything

Pick numbers (or 'skip'):

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.