Skip to content

All versions since 7.0.0

7.0.0

May 11, 2026

Plugin Restructure & Meeting Hygiene

The bloated catalyst-pm plugin splits into four focused tools: catalyst-pm (12 strategy/PRD skills), catalyst-pm-ops (Linear integration and cadence), catalyst-meeting-hygiene (agenda through feedback workflows), and catalyst-discovery (research and prototyping). Skills moved to new plugin paths like /catalyst-pm-ops:daily-plan and /catalyst-meeting-hygiene:meeting-notes, so update any hardcoded references in your workflows.

PRs

  • pm: shrink catalyst-pm to 12 strategy/PRD skills (CTL-322) (#543)
  • pm: extract meeting workflow skills into catalyst-meeting-hygiene plugin (CTL-320) (#541)
  • pm-ops: extract Linear/cadence/comms skills from catalyst-pm (CTL-319) (#537)
  • pm-ops: extract Linear/cadence/comms skills from catalyst-pm (CTL-319) (#537) (5a9c582)
  • pm: extract meeting workflow skills into catalyst-meeting-hygiene plugin (CTL-320) (#541) (ba0c12a)
  • pm: shrink catalyst-pm to 12 strategy/PRD skills (CTL-322) (#543) (43b1c89)

7.1.0 (2026-06-03)

Features

  • pm: AI-native estimation tooling — corpus extractor, calibrated scorer, reference-class lookup (CTL-746) (#1265) (4f56000)

7.2.0 (2026-06-08)

Features

  • pm: close the estimation feedback loop — recurring corpus refresh from real actuals (CTL-813) (#1400) (f066c82)

7.3.0

Jun 30, 2026

Evidence-Based Linear Read Path

Agents now follow a single canonical rule for reading Linear: on standard nodes they call linearis directly, and on Catalyst Cloud nodes they read the replica first and only escalate to a direct read if there’s concrete evidence of staleness. This eliminates the scattered, conflicting guidance that was pointing agents at the retired read mirror. No migration needed — the daemon already handles mode resolution automatically.

PRs

  • dev: CTL-1392 — agents read the Catalyst Cloud replica first (evidence-based fallback) (#2474) (5599fe9)

7.4.0

Jul 01, 2026

Enforced Linear Read Path

The linearis skill now mandates that all agent and script Linear reads go through catalyst-linear read|list|search instead of calling linearis directly. This prevents bare reads from bypassing the local replica and burning shared API quota on rate-limited workers. Scripted read swaps for the .sh helpers and phase skills land in a follow-up PR.

PRs

  • dev: CTL-1397 (1/n) — mandate Linear reads via catalyst-linear (linearis skill + research agents) (#2496) (27ab1fe)

7.5.0

Jul 02, 2026

Direct SQLite Linear Reads

Agents now query the local Catalyst Cloud replica (~/catalyst/catalyst-replica.db) directly with SQL instead of routing through the catalyst-linear CLI wrapper, giving them full access to issues, labels, and relations in a single query — including joins the CLI couldn’t do. A freshness gate mirrors the daemon’s own liveness checks, and any stale or missing replica triggers a loud fallback to linearis rather than a silent reroute. The catalyst-linear CLI is now deprecated and kept only as a compatibility shim.

PRs

  • dev: CTL-1397 — pivot agent Linear reads to direct SQLite; deprecate catalyst-linear CLI (#2514) (249c9d6)

7.6.0 (2026-08-04) Latest

Features

  • dev: CTL-1628 catalyst-runtime-root resolver — fold Tier 1 duplicates + 4 latent resolver bugs (Phase A2) (#2946) (c01ef76)

Bug Fixes

  • pm: CTL-1616 complete PR3 — the score-tickets.ts fold missed the #2919 commit (#2921) (7a31c7a)
  • pm: CTL-1616 harden the score-tickets contract loader (#2921 post-merge Codex x2) (#2922) (6324535)