v1.1.0
NationalSecurityAgency/ghidrav1.1.0May 21, 2026by github-actions[bot]
Upgrading?
AI Summary
This release adds a dedicated Codex council coordinator, NVIDIA NIM provider support, and several deliberation-quality improvements. Round 2 now anonymizes peer outputs and includes an anti-conformity directive, while synthesis is assigned to an explicitly named Chairman model. Verdicts gain actionability sections and a session metadata block, backed by new CI hardening.
Key Highlights
- Dedicated Codex council coordinator plus Codex install support with reliability hardening
- NVIDIA NIM provider support via a new `openai_compatible_api` archetype with runtime credential resolution
- Round 2 cross-examination anonymization masks peer outputs behind stable labels in full and quick modes
- Anti-conformity directive requires members to name the specific flaw in their earlier argument before updating
- Explicit Chairman role selected via STEP 1.7 with a `--chairman` flag; the Chairman cannot be a panel member
Breaking Changes
- Synthesis is now performed by a named Chairman model that cannot simultaneously serve as a panel member
New Features
- Verdict actionability sections: Acceptable Compromises, Kill Criteria, and Concrete Next Step
- Session Metadata block appended to every verdict with best-effort token and duration estimates
- `openai_compatible_api` dispatch path extensible to Together, Fireworks, and vLLM
- CI hardening: `.gitattributes` LF normalization, lint workflow with shellcheck and markdownlint, release workflow with auto-generated notes
- README refresh with header image, quickstart, and open-source best practices
Full Release Notes
### Added - — dedicated Codex council coordinator - Codex install support in (, flags) with reliability hardening - NVIDIA NIM provider support () — auto-detection via - Round 2 cross-examination anonymization — peer Round 1 outputs are masked behind stable labels in full and quick modes (Choi et al., arXiv:2510.07517; Karpathy ) - Anti-conformity directive — Round 2 prompts in all three modes now require members to name the specific flaw in their earlier argument before updating; defends correct prior positions against social pressure (Cui et al., Free-MAD, arXiv:2509.11035) - Explicit Chairman role — synthesis (STEP 7, QUICK STEP 3, DUO STEP 4) is performed by a named model selected via STEP 1.7; new flag, block in , and a hard constraint that Chairman cannot be a panel member - Verdict actionability sections — , , are now required in every verdict (full mode); quick and duo modes require subsets per the per-mode policy - provider archetype — fifth dispatch path alongside subagent / codex_exec / gemini_cli / ollama_run; routes NIM seats (and future Together / Fireworks / vLLM) via with credentials resolved from at runtime - Session Metadata block () — appended to every verdict with , , , , , , plus best-effort token / duration estimates - CI hardening — for LF normalization, (shellcheck + markdownlint), (tarball + auto release notes from CHANGELOG on tag), itself ### Changed - README updated with header image, quickstart, and open-source best practices ### Fixed - Dead code in == Council Simulation Checklist == [PASS] SKILL.md exists [PASS] SKILL.codex.md exists [PASS] Agent definitions found (count=18) [PASS] Provider/model slot template exists [PASS] exploration-orthogonal profile documented in SKILL.md [PASS] execution-lean profile documented in SKILL.md [PASS] --models flag documented in SKILL.md [PASS] --quick mode documented in SKILL.md [PASS] --duo mode documented in SKILL.md [PASS] Execution checkpoints present in SKILL.md [PASS] Verification steps present in SKILL.md [PASS] Round 2 anonymization wired in SKILL.md [PASS] Round 2 anonymization wired in SKILL.codex.md [PASS] Anti-conformity directive present in all 3 Round 2 prompts in SKILL.md [PASS] Anti-conformity directive present in SKILL.codex.md [PASS] Chairman role wired in SKILL.md (STEP 1.7 + --chairman flag + synthesis step) [PASS] Chairman role wired in SKILL.codex.md [PASS] Chairman defaults configured in auto-route-defaults.yaml [PASS] Verdict actionability sections present in SKILL.md (Acceptable Compromises / Kill Criteria / Concrete Next Step) [PASS] Verdict actionability sections present in SKILL.codex.md [PASS] openai_compatible_api archetype wired in SKILL.md (dispatch + base_url + api_key_env) [PASS] openai_compatible_api archetype wired in SKILL.codex.md [PASS] Session Metadata schema wired in SKILL.md [PASS] Session Metadata referenced in SKILL.codex.md [PASS] All agents have consistent section structure [PASS] Grounding protocols placed before Analytical Method in all agents [PASS] All triad member agent files present [PASS] Verdict template dedup check done [PASS] detect-providers.sh exists and is executable [PASS] detect-providers.sh produces valid JSON [PASS] Auto-route defaults config exists [PASS] --no-auto-route flag documented in SKILL.md [PASS] --dry-route flag documented in SKILL.md [PASS] All agents have provider_affinity in frontmatter [PASS] shellcheck passed for install.sh [PASS] install.sh --dry-run completed [PASS] install summary output present [PASS] install.sh --dry-run --copy-configs completed [PASS] config summary output present [PASS] install.sh --dry-run --codex completed [PASS] Codex install summary output present Checklist complete. that tripped ShellCheck SC2317 under CI
Related Repositories
krahets/hello-algo★ 130,899
《Hello 算法》：动画图解、一键运行的数据结构与算法教程。支持简中、繁中、English、日本語，提供 Python, Java, C++, C, C#, JS, Go, Swift, Rust, Ruby, Kotlin, TS, Dart 等代码实现
iluwatar/java-design-patterns★ 94,813
Design patterns implemented in Java
OtterMind/Chat2DB★ 28,329
Chat2DB is a free, cross-platform, local-first database client and SQL workspace for developers, DBAs, analysts, and data teams. Connect to 40+ databases, manage data, edit and run SQL, and use your own AI model to generate, explain, and optimize queries. Available on desktop, web, Docker, and CLI, with MCP support.