v0.3.0

What's Changed

• Fix argo memory and images by @yiyaw-lab in #8
• Public-release readiness: archive, evidence, docs, security by @yiyaw-lab in #9
• Argo: real memory (seen-store persistence), no phantom sends, reply context, tool telemetry by @yiyaw-lab in #10
• Frontier-evolution loop: Argo watches the frontier and proposes its own upgrades by @yiyaw-lab in #11
• Add conftest.py so the test suite can be collected by @yiyaw-lab in #12
• argo: GPT fallback keeps tool access when Claude is down by @yiyaw-lab in #13
• Add Together AI blog feed by @yiyaw-lab in #14
• argo_store: make save_json atomic (temp file + os.replace) by @yiyaw-lab in #15
• argo_mcp_server: claim pending heal atomically (no double-apply) by @yiyaw-lab in #16
• argo_scheduled: run reflect in the local loop so self-reflection fires by @yiyaw-lab in #17
• argo_paths/argo_mcp_server: put the pending-heal store on the volume by @yiyaw-lab in #18
• grok_search/argo_watch: strip XAI key at read + log the silent-disable by @yiyaw-lab in #19
• CLAUDE.md: correct the module-size note by @yiyaw-lab in #20
• argo: enable the frontier loop, add the capability-gap proposer, finish path centralization by @yiyaw-lab in #21
• docs: README hero banner + stats header + self-improvement-loops section by @yiyaw-lab in #22
• Fix propose_change crash + bluffing, chat amnesia, and project 001 reset by @yiyaw-lab in #23
• argo: structured outputs for diagnosis + make EVOLVE drafting land PRs by @yiyaw-lab in #24
• argo: ground the Rehearse verdict in the belief graph + a calibration number by @yiyaw-lab in #25
• argo: reuse shared-utils (tls_context in set_webhook, argo_store in probes) by @yiyaw-lab in #26
• docs: refresh stale README stats + correct propose-token scope by @yiyaw-lab in #27
• argo: load_bets_log fails loud and safe on a corrupt ledger by @yiyaw-lab in #28
• docs: add social-preview card + sync test count to 407 by @yiyaw-lab in #29
• argo: auto-check PR code-review findings + a tool to read them by @yiyaw-lab in #31
• argo: read whole modules + propose surgical edits to large files by @yiyaw-lab in #32
• Argo evolution: adopt prompt_caching by @yiyaw-lab in #30
• argo: EVOLVE/FIX authors surgical edits, not whole-file rewrites by @yiyaw-lab in #33
• docs(CLAUDE.md): proactive (state × operation) matrix-audit guardrail by @yiyaw-lab in #34
• docs(CLAUDE.md): grep every home of a source-of-truth mismatch by @yiyaw-lab in #35
• docs: volume vars + Actions-secret requirement for acted-on-push by @yiyaw-lab in #40
• argo: make the DEFAULT-profile fallback loud (F2a) by @yiyaw-lab in #39
• argo: instrument acted-on-push (F1) via an Actions→Railway HTTP bridge by @yiyaw-lab in #36
• argo: inject a factual persistent-context block into Argo's prompt (F2b) by @yiyaw-lab in #37
• argo: mine the chat log for weakness signals into the diagnose loop (F3) by @yiyaw-lab in #38
• argo: Batch API primitive for bulk SEAS scoring (3rd Stage-1 seed) by @yiyaw-lab in #43
• argo: F7 escalation broker (ask_owner / get_owner_answers) on /mcp by @yiyaw-lab in #45
• argo: F6 rare/right/steerable proactiveness gate on unprompted pushes by @yiyaw-lab in #44
• argo: usage/cost telemetry lever (record per-call tokens + model) by @yiyaw-lab in #41
• docs: correct stale CLAUDE.md seam note + name the real argo_webhook seam by @yiyaw-lab in #46
• argo: cost-based prediction kind (cache_ratio) -- first MEASURED cost claim by @yiyaw-lab in #47
• argo: wire the Batch primitive into SEAS scoring (opt-in, dry-run gated) by @yiyaw-lab in #50
• argo: extract argo_bluff + argo_media seams from argo_webhook (pure refactor) by @yiyaw-lab in #51
• argo: ambient status (H3.3) + who-acts-next classifier by @yiyaw-lab in #49
• argo: F5 receipts -- surface Argo's graded track record, honestly by @yiyaw-lab in #48
• argo: extract F7 escalation broker into argo_escalation (pure refactor) by @yiyaw-lab in #52
• argo: score SEAS findings' dated predictions (finding_prediction metric) by @yiyaw-lab in #53
• argo: CMO role lens (B-007 cheapest demand test) by @yiyaw-lab in #54
• docs: document the 4 volume vars added since #40 (CMO lens + siblings) by @yiyaw-lab in #55
• seas: depth signals (architecture diagram, CI badge, CONTRIBUTING) by @yiyaw-lab in #57

Full Changelog: https://github.com/yiyaw-lab/seas/commits/v0.3.0