v2.2.0

Added

• A skills-only Codex package for refine-gate, exposed through a repository
  marketplace at .agents/plugins/marketplace.json.
• Gemini CLI installation through the existing portable Agent Skill.
• OpenSSF Scorecard, CodeQL, Dependabot, and a hash-locked development
  dependency set.
• OpenSSF Best Practices Passing evidence and its repository badge.
• Security, contribution, conduct, governance, architecture, assurance-case,
  Scorecard, and twelve-month roadmap documentation.
• A release workflow that tests and self-verifies a source bundle, publishes
  SHA-256 checksums, an executable manifest and CycloneDX SBOM, and creates
  Sigstore build-provenance attestations.
• Regression tests for context-guard hooks, prompt-refinement measurement,
  statusline transcript handling, and release integrity.

Changed

• The refine skill now uses the portable Agent Skills frontmatter and
  host-neutral wording. Claude's UserPromptSubmit hook and plugin manifests
  are unchanged.
• CI actions are pinned to full commit SHAs, workflow permissions are read-only
  by default, ShellCheck is checksum-verified, and Python dependencies are
  installed from the hashed lock file.
• CI measures the shipped Python surface with coverage.py's subprocess support
  and enforces an 80% floor; the initial complete measurement is 94%.
• Python contributions now name PEP 8 plus Ruff's selected rules as the coding
  standard, and CI enforces those rules from the hash-locked toolchain.
• The repository introduction now leads with its portable Codex, Gemini CLI,
  Claude, and Agent Skills surface while preserving explicit labels on
  Claude-only integrations.

Fixed

• The statusline transcript scanner now recognizes valid compaction records
  whose JSON contains insignificant whitespace before confirming the parsed
  marker values.

v1.4.3 — DS gate G6 heat-track fix

What's Changed

• fix(statusline): 4-palier heat track, replaces continuous grad_rgb lerp by @cdeust in #6
• Release 1.4.3 by @cdeust in #7

Full Changelog: v1.4.2...v1.4.3

v1.4.2 — statusline conforme AI Architect DS

What's Changed

• design(statusline): palette AI Architect DS (remplace Catppuccin Mocha) by @cdeust in #2
• design(statusline): AI Architect DA compliance — neutral chrome, single accent by @cdeust in #3

Full Changelog: v1.4.1...v1.4.2

v1.4.1 — AI Architect banner

What's Changed

• design(session-optimizer): bannière AI Architect by @cdeust in #1

New Contributors

• @cdeust made their first contribution in #1

Full Changelog: v1.4.0...v1.4.1

v1.4.0 — subagent usage tracker + enriched statusline

session-optimizer v1.4.0

Deux ajouts majeurs réunis sous cette release.

🤖 Subagent usage tracker (SubagentStop hook)

• hooks/subagent-tracker.py + tools/subagent_usage.py : comptage réel des tokens/coût dépensés par les subagents sur la session (ce que l'input statusline omet).
• hooks/stop-context-guard.py enrichi ; tests tests/test_subagent_usage.py.

📊 Statusline enrichi

• Git : ahead/behind, stash, état working tree (↑/↓/⚠/!M+A✘D?U).
• Télémétrie transcript (statusline-transcript.py) : débit ⚡ t/s, âge de la dernière réponse 🕑, compte de compactions 🗜, compte à rebours du cache de prompt ❄ (TTL 5 min).
• Jauges quota 🎯 5h / 7d mappées sur les fenêtres de rate-limit Pro/Max (la vraie contrainte « ne pas dépasser ») + référence de coût 💰 $/mois et 🤖 $/run.
• Companions : statusline-costs.py, statusline-budget.json (config verbosité + TTL, aucun budget absolu), README-statusline.md.

Compat

Rétro-compatible. Les hooks tournent en CLI, app desktop, extensions IDE et Cowork. Surface chat claude.ai : pas de hooks — utiliser skills/refine/ comme Agent Skill.