v1.8.0

release: v1.8.0

2026 X API alignment release. Highlights:
- official Articles endpoints: 'article draft --api' + 'article publish'
- Apr 20 API removals surfaced honestly (likes/follows/quote writes),
  quote posts and >280 Premium posts fall back to the web session
- HTTP 402 pay-per-use errors mapped with a console.x.com hint
- reply-back tracking and metrics deltas fixed on legacy DBs (were
  silently dead), auto-schedule timezone fix, double-post claim token
- --csv output wired, May 2026 search operators documented
- fabricated algorithm weight claims removed from the agent skill

Claude-Session: https://claude.ai/code/session_01UDsdErgNhSnKsRPXD4ubvn

v1.7.5

release: v1.7.5

Internal refactor and one bug fix; no user-facing behaviour change.

- fix(schedule): truncate previews on char boundaries (no more panic on
  multi-byte/emoji content in schedule list/add/fire)
- refactor: remove dead timing-analytics methods and an unused wrapper
- refactor(xapi): fold engagement verbs, share the pagination loop and
  tweet-listing tail, drop a throwaway clone in get_tweet
- refactor(errors): convert rusqlite errors via From in the tracker
- style: clear the items-after-test-module clippy warning

Claude-Session: https://claude.ai/code/session_01TXjeyZcV3Qr6CozGBmCi7s

v1.7.4

v1.7.4: recency hardening — search age + 24h floor; engage feed 30-mi…

v1.7.3

v1.7.3: fix install.sh asset names; auto-publish to crates.io + auto-…

v1.7.2

v1.7.2 — fix opt-out flag rejection + bad-ID silent fallback (v1.7.1 …

v1.7.1

v1.7.1 — FxTwitter Article enrichment

Closes the Article-blindness gap: xmaster read/analyze now auto-resolve
X Article bodies via FxTwitter (no auth, soft dep, ~2.5s timeout, opt-out
via settings.disable_fxtwitter). xmaster analyze accepts tweet IDs and X
URLs directly, fetches the content, and scores Article bodies — not the
t.co wrapper. Hashtag counter fixed to skip markdown headers in long-form
bodies.

v1.7.0

v1.7.0 — integration patch + research-derived follow-conversion gates

Embeds the multi-agent follow-conversion playbook into preflight, config
check, engage recommend, inspire, and metrics. Bakes in the rules so users
don't have to memorize them.

New preflight gates: weak_hook_no_weight (Hook), reference_only_risk
(Substance) — catches the doublenickk trap, multiple_ctas (Substance) —
Welsh's one-ask rule. Issue now carries a group field for action-oriented
UX without breaking the JSON contract.

Bio audit in config check scans against the 4-element follow-conversion
formula. inspire freshness annotation prevents near-duplicate reposts.
metrics adds profile_click_rate_per_1k + bookmark_to_like_ratio as
leading follow-conversion indicators. engage recommend deprioritizes
sub-500-follower targets (Grox spam-classifier bucket). Home-timeline
endpoint URL bug fixed.

Zero new commands. Zero new flags. ~440 LOC across 9 files. 95 unit + 29
integration tests pass; clippy -D warnings clean.

v1.6.8

v1.6.8 — cut historical / empirical / fabricated bloat

Schema break (clean cut, not deprecation): drop agent_info.algorithm.weights
in favor of agent_info.algorithm.signals (name + polarity). Drop
time_decay_halflife_minutes, out_of_network_reply_penalty, and 3 of 5
EngagementHeuristic fields. Trim all 'historical 2023 estimate' and
'empirical, not in code' framing from issue messages and usage hints.
Output size -33%, codebase -82 LOC net.

v1.6.7

v1.6.7 — algorithm honesty release + 4 follow-up fixes

Strip 2023-leak claims that were not present in the May 15 2026 xai-org/x-algorithm
re-release. Add 3 missing scorer terms (not_dwelled, quoted_vqv, cont_click_dwell_time).
Replace fabricated weight magnitudes with honest 'historical 2023 estimate' framing.
Fix 'track status' SQLite TEXT-affinity crash and home-timeline null-body crash.
5 new regression tests prevent reintroduction of the false claims.

Detailed audit: 5 independent agents (3 Claude Opus + 2 Codex GPT-5.5)
reviewed xmaster against the May 15 2026 source. GPT-5.5 Pro meta-reviewed
the findings. Verified end-to-end by 5 more subagents before release.

xmaster v1.6.6

Adds xmaster engage inbox, a post-launch conversation audit that checks root replies, quote tweets, and replies under quote tweets in one pass. It ranks likely reply-back opportunities using the historical Twitter open-model reply_engaged_by_author heuristic while clearly caveating that live X weights may differ.