Skip to content

CHANGELOG.md

Latest commit

 

History

History
78 lines (62 loc) · 4.87 KB

CHANGELOG.md

File metadata and controls

78 lines (62 loc) · 4.87 KB

Changelog

v2.3.0 (2026-02-09)

Fixed — Remove LLM Hallucinations

Most LLMs have the old X API tier system (Basic/Pro/Enterprise, $200/mo subscriptions) baked into their training data. This caused confusion for users whose agents referenced pricing and access levels that no longer exist. This release updates all skill docs to reflect the current pay-per-use model so your agent has accurate information.

  • Purged all stale tier/subscription references across 6 files (13 instances of "Basic tier", "current tier", "enterprise-only" etc.)
  • Full-archive search (/2/tweets/search/all) is available on pay-per-use — not enterprise-only as LLMs commonly claim
  • Updated rate limits — old per-15-min caps replaced by spending limits in Developer Console
  • Clarified 7-day limit is a skill limitation (using recent search endpoint), not an API restriction
  • Updated query length limits — 512 chars (recent), 1024 (full-archive), 4096 (enterprise)
  • Added per-resource cost breakdown: $0.005/post read, $0.010/user lookup, $0.010/post create
  • Added 24-hour deduplication docs, xAI credit bonus tiers, usage monitoring endpoint

Fixed

  • Tweet truncation bug — tweet and thread commands now show full tweet text instead of cutting off at 200 characters. Search results still truncate for readability. (h/t @sergeykarayev)

Added

  • Security section in README — Documents bearer token exposure risk when running inside AI coding agents with session logging. Includes recommendations for token handling.

v2.2.0 (2026-02-08)

Added

  • --quick mode — Smarter, cheaper searches. Single page, auto noise filtering (-is:retweet -is:reply), 1hr cache TTL. Designed for fast pulse checks.
  • --from <username> — Shorthand for from:username queries. search "BNKR" --from voidcider instead of typing the full operator.
  • --quality flag — Filters out low-engagement tweets (≥10 likes). Applied post-fetch since min_faves operator isn't available via the API.
  • Cost display on all searches — Every search now shows estimated API cost: 📊 N tweets read · est. cost ~$X

Changed

  • README cleaned up — removed duplicate cost section, added Quick Mode and Cost docs
  • Cache supports variable TTL (1hr in quick mode, 15min default)

v2.1.0 (2026-02-08)

Added

  • --since time filter — search only recent tweets: --since 1h, --since 3h, --since 30m, --since 1d
    • Accepts shorthand (1h, 30m, 2d) or ISO 8601 timestamps
    • Great for monitoring during catalysts or checking what just dropped
  • Minutes support (30m, 15m) in addition to hours and days
  • Cache keys now include time filter to prevent stale results across different time ranges

v2.0.0 (2026-02-08)

Added

  • x-search.ts CLI — Bun script wrapping the X API. No more inline curl/python one-liners.
    • search — query with auto noise filtering, engagement sorting, pagination
    • profile — recent tweets from any user
    • thread — full conversation thread by tweet ID
    • tweet — single tweet lookup
    • watchlist — manage accounts to monitor, batch-check recent activity
    • cache clear — manage result cache
  • lib/api.ts — Typed X API wrapper with search, thread, profile, tweet lookup, engagement filtering, deduplication
  • lib/cache.ts — File-based cache with 15-minute TTL. Avoids re-fetching identical queries.
  • lib/format.ts — Output formatters for Telegram (mobile-friendly) and markdown (research docs)
  • Watchlist system — data/watchlist.json for monitoring accounts. Useful for heartbeat integration.
  • Auto noise filtering — -is:retweet added by default unless already in query
  • Engagement sorting — --sort likes|impressions|retweets|recent
  • Post-hoc filtering — --min-likes N and --min-impressions N (since X API doesn't support these as search operators)
  • Save to file — --save flag auto-saves research to ~/clawd/drafts/
  • Multiple output formats — --json for raw data, --markdown for research docs, default for Telegram

Changed

  • SKILL.md rewritten to reference CLI tooling. Research loop instructions preserved and updated.
  • README.md expanded with full install, setup, usage, and API cost documentation.

How it compares to v1

  • v1 was a prompt-only skill — Claude assembled raw curl commands with inline Python parsers each time
  • v2 wraps everything in typed Bun scripts — faster execution, cleaner output, fewer context tokens burned on boilerplate
  • Same agentic research loop, same X API, just better tooling underneath

v1.0.0 (2026-02-08)

Added

  • Initial release
  • SKILL.md with agentic research loop (decompose → search → refine → follow threads → deep-dive → synthesize)
  • references/x-api.md with full X API endpoint reference
  • Search operators, pagination, thread following, linked content deep-diving