Index of every page on this site: llms.txt
All skills
kunchenguid avatar

/whathappened

@fe9e84b
by Kun Chenkunchenguid/whathappened229 stars
17
Run it yourself

Brief what just happened on a topic and map public opinion from X (Twitter) only. Adaptive recency window biased to the freshest coherent story. Soft exception: at most one web lookup for entity resolution. Use when the user runs /whathappened, or asks what happened / what's going on / X or Twitter reaction / public opinion on X about a person, product, launch, or event. Grok Build only (requires native X tools). Neutral analyst tone. No discovery mode in v1.

Use this Skill: https://skilld.dev/gh/kunchenguid/whathappened/whathappened

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈119 tokens for metadata: the name and description. ≈2k when used: this file. ≈1.5k more on demand in 2 files.

Description uses 6.0% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈478 characters, or 6.0%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

/whathappened

You are running the whathappened skill. Produce a neutral briefing of what happened and what public X is saying about a named topic.

This is not generic research. Follow this file top to bottom.

Host requirement

You need these tools:

  • x_keyword_search
  • x_semantic_search
  • x_thread_fetch
  • x_user_search

If they are missing, stop. Tell the user this skill only works in Grok Build (or another host with those X tools). Do not fake an X briefing from web search.

Hard rules

  1. X-first synthesis. All claims about the event and public opinion come from X tool results. Never invent posts, handles, engagement, or quotes.
  2. Soft web exception. At most one web lookup (web_search or web_fetch) only to resolve what/who the topic is (official name, handle, product identity). Do not use web for opinions, sentiment, or "what people think." If you use the web lookup, say so in Gaps.
  3. No discovery. If the user gives no topic, or asks "what's trending / what's hot on X," do not invent a feed. Ask for a topic in one short question.
  4. No flags in v1. Ignore or gently ignore flag-like tokens (--deep, window overrides). Infer window from the pulse (below).
  5. Neutral analyst tone. No hype voice, no dunking, no "the room is screaming." Report camps and debates as observed.
  6. Adaptive window, freshest-first. Do not default to 30 days. Prefer the shortest window that still explains the story.
  7. Opinion is not ground truth. Frame sentiment as public conversation on this X sample.
  8. Citations. Prefer @handle plus post links when available. Use the host's inline post citation render when you have post citation IDs from X tools. Never invent URLs. Never append a trailing dump of unrelated Sources.

Read references/query-patterns.md when building search queries. Read references/failure-modes.md when the sample looks thin, mixed, or noisy.

Pipeline

Step 0 - Parse topic

  • Require a named topic (person, org, product, launch, event, controversy).
  • Strip intent modifiers for search entities (sentiment, drama, takes) but keep them as synthesis focus if the user asked for them.
  • Classify loosely: event | person/org | product/release | debate.

Step 1 - Optional entity web lookup (0 or 1 call)

If the topic is ambiguous or you lack an obvious official handle/name:

  • Make at most one web lookup to resolve identity.
  • Extract: canonical name, official @handle if stated, aliases, event date if the page states one.

If the topic is already clear (e.g. Kimi K3, @sama), skip web.

Step 2 - X entity grounding

  • Use x_user_search for likely official or primary accounts when useful.
  • Build aliases: exact phrase, alternate spellings, product codes, cashtags.
  • Prefer first-party from:handle when you have a confident handle.

Step 3 - Pulse (always first)

Run a cheap Latest keyword search with a short since: window to measure velocity. Example windows to try first:

  • Breaking candidate: last few hours (since:YYYY-MM-DD for today, Latest)
  • If almost empty: widen once (yesterday / last 2 days) before committing mode

Commit a mode and window:

Pulse signal Mode Default window
High volume, rising engagement, clear origin Breaking last 1–6 hours (minutes if still exploding)
Clear same-day event / launch chatter Same-day last 24–48 hours
Ongoing debate, moderate velocity Story last 3–7 days
Sparse posts, mostly references to older news Background last 14–30 days; still prefer recent posts

State the chosen Window and Mode in the brief. Expand later only if:

  • No agreement on what happened
  • Missing first-party post from the obvious account
  • Debates point at an origin post you have not seen

When ranking, prefer recency over raw likes unless an older post is clearly the origin everyone quotes or replies to - then fetch that thread.

Step 4 - Search lattice (parallel)

Use multiple lanes. Cap total X tool calls roughly 8–14 for a normal run.

Lane Tool Goal
Top x_keyword_search mode=Top High-engagement consensus
Latest x_keyword_search mode=Latest + time operators Fresh narrative
Semantic x_semantic_search Paraphrases and adjacent framing
First-party keyword from:handle when known Official statement
Debate controversy / quote / counter-claim queries Disagreement surface

Use advanced operators from references/query-patterns.md. Always keep the primary entity in the query. Drop off-entity viral noise.

Step 5 - Thread enrichment

x_thread_fetch the 3–8 highest-signal posts, prioritizing:

  1. Origin / announcement
  2. Official first-party
  3. Highest-engagement summary take
  4. Strongest steelman and strongest criticism
  5. Posts that define a camp split

Step 6 - Rank and cluster (in head, lightweight)

Rough score:

score = engagement × freshness_weight × authority_weight × on_entity
  • On-entity: must clearly be about the primary topic (or hard alias).
  • Authority (soft): official accounts, domain experts, primary reporters - never a hard allowlist; do not over-weight bluechecks alone.
  • Clusters: merge posts that make the same claim into one camp bullet.

Discard engagement bait that fails entity grounding.

Step 7 - Synthesize

Emit the brief using the template below. Do not dump raw ranked lists. Include at least 2–3 short attributed quotes from real posts when the sample has usable text.

If the sample is thin, say so. Prefer honest uncertainty over fake consensus.

Output template

Use this structure. Keep headings. No extra blog-style sections.

# /whathappened: {topic}

**Window:** {human window} · mode {Breaking|Same-day|Story|Background} · as of {UTC or local stamp}
**X sample:** ~{N} posts · Top + Latest + {K} threads · confidence {high|medium|thin}
**Entity resolve:** {none | one web lookup: one-line what it was for}

## What happened
2–5 sentences. Prefer first-party and origin posts. Stick to what X supports.

## Where the conversation is
- Dominant frame
- Secondary frames
- Who is loud (camps / account types), not "people in general"

## Public opinion map
| Camp | Share (rough) | Core claim | Representative voices |
|------|---------------|------------|------------------------|
| … | ~X% | … | @a, @b |

Label shares as qualitative judgment from this sample, not polling.

## The live debates
1. **{title}** - Side A vs Side B; stakes in one line; best evidence posts
2. …

## Notable posts (receipts)
- [@handle](url or status link if known) - why it matters (role: origin / official / steelman / critique / meme)
- … (5–10 max)

## Gaps / caveats
- Thin sample, missing official voice, language bias, bot/noise risk, web resolve used, etc.
- What would change this read if true

Formatting notes

  • Use plain hyphens -, not em dashes.
  • Neutral wording: "many posts claim", "a common critique is", not "everyone knows".
  • Rough camp percentages must be marked as rough sample judgment.
  • If confidence is thin, shrink the opinion map and expand Gaps.
  • End after Gaps. No trailing invitation spam unless the user asked follow-ups.

Pre-flight checklist (before first tool call)

  • Topic present (else ask)
  • X tools available (else refuse)
  • Web budget remaining: 0 or 1 resolve-only
  • No discovery request

Post-synthesis checklist (before send)

  • Window + mode stated
  • Every quoted line attributable to a fetched post
  • No web-derived "sentiment"
  • Camps grounded in multiple posts when confidence is high
  • Receipts are real and on-entity
  • Gaps honest about sample limits

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts4mo3 checks · Risk SAFE
  • Gen Agent Trust Hub4mo

    The whathappened skill is a specialized tool for analyzing events and public sentiment on X (Twitter) using native Grok Build tools. The skill implements strict logic for source attribution, neutral reporting, and entity resolution with a minimal web search footprint. No malicious patterns, obfuscation, or unauthorized data access were detected.

  • Socket4mo

    No alerts

  • Snyk4mo

    Risk: LOW · No issues

Signed by skilld at fe9e84b. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last week.

Steadyupdated 4 months ago
metadata
{
  "short-description": "X-only adaptive briefing of events + public opinion"
}

README badge

README badge for kunchenguid/whathappened