v4.2.0

Summary

• Hardened API-primary Linear execution with required MCP fallback when API execution blocks or read-back cannot prove state.
• Added expanded-mode outer-harness research, decisions, QA, dependency map, and recursive loop artifacts.
• Added v4.2.0 evaluator coverage for fallback wording and dogfood attachment read-back.

Verification

• Local regression tests and Python graph tests passed.
• Front-door, agent, audit, and dogfood evaluators passed.
• GitHub PR checks passed before merge.

v4.1.0 - Reactive Linear issue spillover

Highlights

• Adds reactive Linear issue spillover for direct GraphQL graph apply when Linear rejects an issue description for size or character limits.
• Adds --spillover-dir and project.spilloverDir so spillover docs can be routed to a known local context root.
• Keeps normal issues lightweight: Linear is tried first, and local per-issue docs are only created after an actual Linear rejection.
• Adds fake Linear limit coverage, read-back tolerance for spillover pointer bodies, and dogfood fake-state setup.

Verification

• CI test: passed
• CodeQL analyze: passed
• Local shell syntax, Python tests, evaluators, temp install check, README completeness, and publisher secret preflight passed.

v4.0.1

• Made direct Linear GraphQL the documented primary path whenever LINEAR_API_KEY or LINEAR_ACCESS_TOKEN is available; Linear app/MCP/manual actions are now explicitly fallback-only unless requested.
• Updated dry-run transition output so agents see the direct API path first and fallback connector actions second.
• Scoped graph-apply title idempotence to the target project to avoid reusing generic guide issue titles from unrelated Linear projects.
• Added Context7/official Linear docs checkpoint coverage to the v4.0.1 project and evaluator coverage for API-primary routing.

v4.0.0

• Aligned live GraphQL mutations with Linear's current schema by replacing the
  stale project teardown mutation with projectDelete.
• Added issueBatchCreate for efficient graph bootstrap while preserving
  idempotent lookup/update behavior and per-issue read-back verification.
• Upgraded live project read-back and project-scan reconciliation to follow
  cursor pagination across all project issues at 250 issues per page.
• Expanded direct graph support for related and duplicate issue relations,
  richer attachment metadata, and attachment read-back drift detection.
• Added header-aware rate-limit budget tracking on top of the existing
  RATELIMITED / HTTP 429 exponential backoff path.
• Re-pinned the frozen CLI evaluator for the v4 API contract and refreshed
  documentation for the direct API mode.

v3.6.0

• Lifted the live graph-apply --apply-linear gate so project graphs can be
  created directly through Linear's GraphQL API without requiring a Linear MCP
  server for project creation.
• Added idempotent direct GraphQL operations for projects, labels, milestones,
  issues, issue relations, and attachments, with fake transport handlers,
  rate-limit retries, mutation success checks, and read-back verification.
• Added live graph read-back and project-scan reconciliation so the CLI can
  detect Linear issues missing from the ledger and ledger rows missing from
  Linear.
• Added linear-agent discover and linear-agent promote for Continuous Issue
  Discovery rows and optional live Linear issue creation.
• Updated live Linear smoke coverage, docs, evaluator expectations, and the ADR
  trail for direct API mode. Linear MCP remains a fallback for runtimes without
  credentials, not the graph-creation source of truth.

v3.5.0 - Dogfood validation and source checkpoints

Summary

• Added the live dogfood evaluator and five sandbox plan fixtures for small, medium, large, expanded-mode, and continuous-discovery planning.
• Added source material checkpoint guidance covering kickoff, planning, issue creation, implementation, delegation, scope changes, verification, handoff, and closeout.
• Extended fake Linear graph apply/read-back behavior so dogfood tests validate issue descriptions, milestones, links, and agent-ready issue quality.
• Updated version metadata, changelog, and publisher release manifest for v3.5.0.

Verification

• python3 tools/linear-dogfood-evaluator.py: DOGFOOD_SCORE 110/110
• bash tests/test-linear-agent.sh: passed
• python3 -m unittest tests/test_linear_agent_graphql.py tests/test_linear_agent_graph_features.py: 20 tests passed
• python3 tools/linear-agent-evaluator.py: SCORE 130/130
• python3 tools/linear-front-door-evaluator.py: SCORE 235/235
• python3 tools/linear-skill-audit-evaluator.py: ALL_VERTICALS 10/10
• README completeness: 100/100
• install health check: passed
• PR #20 checks: CodeQL, analyze, and test passed

Compare: v3.4.0...v3.5.0

v3.4.0 - Continuous discovery planning model

Summary

• Added Continuous Issue Discovery across normal and expanded mode with create/propose/log decision rules.
• Added compact normal-mode project descriptions and issue bodies for context recovery.
• Added expanded-mode project charters and rich issue bodies for long-horizon multi-agent delivery.
• Added Project Principles / Fundamentals surfaces, fixtures, evaluator coverage, and a stricter release gate.

Verification

• GitHub PR #18 checks passed before merge.
• Local regression suite, unit tests, evaluators, temp install check, and release metadata check passed before release.

v3.3.0 - Expanded planning mode

Summary

• Adds opt-in expanded mode for long-horizon Linear planning projects.
• Adds references/expanded-mode.md, expanded-mode templates, research/spec/dogfood docs, and README guidance.
• Adds evaluator coverage proving expanded mode is additive and baseline mode remains the default.

Verification

• Local full verification passed, including shell regression, unittest graph tests, front-door evaluator 150/150, CLI evaluator 130/130, all-verticals audit, syntax/YAML checks, install sync, and install check.
• GitHub PR checks passed: test, analyze, and CodeQL.

v3.2.0 - Safe parallelism contract

Highlights

• Added an explicit safe-parallelism contract requiring repeated checkpoints, dependency/write-scope/risk/verification grouping, worktree-safe parallel agents, bounded context-isolation agents, serial-constraint explanations, and follow-up issue splitting.
• Expanded execution references and the companion ledger template with concrete safe-parallelism allocation fields and checkpoint recording.
• Added front-door evaluator and all-verticals coverage for the safe-parallelism contract.

Verification

• GitHub PR #15 checks passed: test, analyze, CodeQL.
• GitHub PR #16 checks passed: test, analyze, CodeQL.
• Local checks passed: shell regression tests, GraphQL/graph feature unit tests, front-door evaluator 125/125, CLI evaluator 130/130, all-verticals evaluator 10/10, syntax/YAML checks, install with docs, installed parity check.
• shellcheck was skipped locally because it is not installed.

v3.1.0 - Front door portability

Highlights

• Slimmed SKILL.md into a trigger-safe front door backed by focused references for project structure, execution hygiene, validation modes, and trigger preservation.
• Added source-backed research on progressive disclosure, metadata-driven discovery, and tool routing for skill activation reliability.
• Added a front-door evaluator and wired it into CI/release verification.
• Clarified Codex-first and Claude-compatible portability, including current manual Claude install caveats.
• Documented the manual, secret-gated live Linear smoke-test pattern separately from hermetic fake-transport CI.

Verification

• GitHub PR #14 checks passed: test, analyze, CodeQL.
• Local release worktree checks passed: shell regression tests, GraphQL/graph feature unit tests, front-door evaluator 110/110, CLI evaluator 130/130, all-verticals evaluator 10/10, syntax/YAML checks.
• shellcheck was skipped locally because it is not installed.