| v0.10.7 | Resolves #188, #189, #190. ### Fixes - **`opena2a login --ci` fails fast** (#189): exits non-zero immediately with an actionable message (pointing at `--api-key`) instead of blocking on `Waiting for authentication...`. A CI job with cached valid credentials still returns 0. - **opena2a-prefixed next-steps** (#190): `trust` / `publish` / `registry --help` and opena2a-cli's own setup hints now cite `opena2a` commands instead of bundled tool names. Delegated ai-trust stdout is rebranded line-buffe | High | 6/3/2026 |
| v0.10.4 | ## What's Changed * ci(release): per-package tag triggers + wire @opena2a/ai-classifier by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/87 * feat(cli-ui): export observations + analyst-render for shared CLI use by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/88 * feat(cli): wire @opena2a/cli-ui renderObservationsBlock into opena2a review by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/89 * fix(cli): aggregate HMA findings into opena2a revi | High | 5/28/2026 |
| telemetry-v0.3.0 | ## What's Changed * docs(readme): mirror AIM README structure by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/146 * fix(protect): anchored CLI self-exemption โ replace substring marker check (closes #77) by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/147 * fix(cli): register secure alias + check --nanomind/--rescan flags (closes #135) by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/141 * feat(credential-patterns): 0.1.1 โ block-comment ma | High | 5/24/2026 |
| cli-v0.10.3 | ## What's Changed * chore(cli): bump to 0.10.3 โ adopt @opena2a/telemetry 0.2.0 by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/145 **Full Changelog**: https://github.com/opena2a-org/opena2a/compare/telemetry-v0.2.0...cli-v0.10.3 | High | 5/11/2026 |
| cli-v0.10.2 | ## Security - **OAuth tokens move from plaintext `~/.opena2a/auth.json` to OS keychain.** macOS Keychain via `security`; Linux Secret Service via `secret-tool`. Service `opena2a-cli`, account `${serverUrl}:access` / `:refresh` โ visible as discrete entries in macOS Passwords.app. Metadata file at `~/.opena2a/auth.json` (mode `0600`) retained with new `tokenStorage: 'keychain' | 'file'` discriminator. Under keychain mode the file does NOT contain the token strings. - **Transparent migration on f | High | 4/30/2026 |
| cli-v0.9.1 | ## What's Changed * fix(cli): wire --no-contribute end-to-end (closes #107) โ 0.9.1 by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/108 **Full Changelog**: https://github.com/opena2a-org/opena2a/compare/cli-v0.9.0...cli-v0.9.1 | High | 4/29/2026 |
| check-core-v0.1.0 | ## What's Changed * feat(check-core): extract @opena2a/check-core 0.1.0 (CA-034 M3) by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/97 **Full Changelog**: https://github.com/opena2a-org/opena2a/compare/cli-ui-v0.3.0...check-core-v0.1.0 | High | 4/23/2026 |
| ai-classifier-v0.1.1 | ## What's Changed * ci(release): per-package tag triggers + wire @opena2a/ai-classifier by @thebenignhacker in https://github.com/opena2a-org/opena2a/pull/87 **Full Changelog**: https://github.com/opena2a-org/opena2a/compare/v0.8.24...ai-classifier-v0.1.1 | High | 4/22/2026 |
| v0.8.23 | ### Bug Fixes - `--server cloud` now resolves to `https://aim.oa2a.org` (AIM Cloud Phase 7 backend). Previously pointed to `api.aim.opena2a.org`, which serves a different product (community). Bare `aim.opena2a.org` still routes to `api.aim.opena2a.org` for community users. - Updated `--server` help text and login error message to reference the new default. ### Install - npm: `npm install -g opena2a-cli@0.8.23` - Homebrew: `brew upgrade opena2a` | High | 4/14/2026 |
| v0.8.22 | ## What's new - **AnaLM flag**: `--analm` global option threaded through adapter system to HMA/ai-trust ## Bug fixes - Fix `scan secure .` routing (no longer errors with "secure does not exist") - Fix `check` error showing `hackmyagent` instead of `opena2a` for unrecognized targets | Medium | 4/13/2026 |
| v0.8.21 | Picks up hackmyagent 0.17.1 (unified scoring, analyzer gating, FP reduction) and ai-trust 0.2.24 (default local scan, error status display). | Medium | 4/12/2026 |
| v0.8.19 | Bump MIN_HMA_VERSION to 0.15.7 for unified publish endpoint support | High | 4/9/2026 |
| v0.8.18 | ## What's New - **Unified publish endpoint**: Scan results now POST to `/api/v1/trust/publish` - Legacy `/api/v1/trust/scan-report` used as fallback on 404 - Fix: `check express` now works (bumped HMA dep to ^0.15.7) - Fix: `train --help` no longer tries Docker delegation - Fix: removed erroneous 'check' from publish defaultArgs | Medium | 4/9/2026 |
| v0.8.16 | ## What's New - **Runtime HMA version check**: Warns when installed hackmyagent is below minimum required version (0.15.6) - Prevents confusing errors when using GitHub repo scanning with old HMA - GitHub target routing from feat/github-check now included | High | 4/9/2026 |
| v0.8.15 | ## opena2a-cli v0.8.15 ### Fixes - `runtime bogus` now shows descriptive subcommand list (matches guard/shield) - `publish --ci` no longer errors with "unknown option '--ci'" - Removed blind `--ci` injection into adapter subprocesses (broke tools that don't support it) | Medium | 4/3/2026 |
| v0.8.14 | ## opena2a-cli v0.8.14 ### Fixes - Forward `--format` and `--ci` flags to adapter subprocesses (scan, check, secrets, broker, etc.) - `opena2a scan --json` now produces JSON output (previously silently ignored) - Guard and runtime no-args now show descriptive subcommand lists | Medium | 4/3/2026 |
| v0.8.13 | ## opena2a-cli v0.8.13 ### Fixes - **guard**: Accept directory as positional arg (e.g. `opena2a guard .`) defaults to `status` subcommand - **shield**: Show subcommand list when run with no args instead of cryptic error - **runtime**: Same fix -- show subcommand help on no args - **--json**: Added as global flag (shorthand for `--format json`) works on all commands - **identity verify**: Fix crash with aim-core 0.1.2 (`unexpected type, use Uint8Array`) - All commands with subcommands now show f | Medium | 4/3/2026 |
| v0.8.12 | Add READMEs for shared/contribute, sync npm READMEs | Medium | 3/25/2026 |
| v0.8.10 | ### Review Dashboard Overhaul - **Real HMA scan** โ no more hardcoded score of 70; runs actual `hackmyagent secure` - **Deduplicated findings** โ HMA: 103K rows โ 30 unique checks with occurrence counts; Credentials: 515 โ 7 grouped types - **Risk education** โ amber "Why this matters" text on each HMA finding, 4-section strategic remediation guide - **Merged Integrity into Hygiene** โ 7 tabs โ 6 (Overview, HMA, Credentials, Shadow AI, Shield, Hygiene) - **Dynamic trust score** โ formula shows a | Low | 3/21/2026 |
| v0.8.9 | ## Trust command fixes - Scan data now displays correctly (status, findings, last scanned date) - Supply chain vulnerability breakdown shown (was "Dependencies: undefined") - Weekly download counts displayed when available - Alternate source fallback: tries npm, pypi, github before giving up - Search fallback: shows 5 similar packages when package not found in registry | Low | 3/20/2026 |
| v0.8.8 | Added @opena2a/contribute shared community trust contribution client. Fixed contribute endpoint to use api.oa2a.org. | Low | 3/19/2026 |
| v0.8.0 | ## OAuth Device Flow Login New commands for browser-based authentication with AIM servers: - `opena2a login` - Authenticate via browser (OAuth Device Authorization Grant, RFC 8628) - `opena2a logout` - Remove stored credentials - `opena2a whoami` - Show authentication status After login, all identity commands automatically use stored tokens -- no more `--api-key` needed. ### How it works ``` opena2a login # opens browser, shows code XXXX-XXXX | Low | 3/18/2026 |
| v0.7.2 | ### UX Fix - Changed 'no identity' to 'project not registered' -- the identity is about the project, not the agent | Low | 3/15/2026 |
| v0.7.1 | ### Bug Fix - **`mcp audit`** -- Fixed trust score endpoint. Was querying `/api/v1/packages/{name}` (404), now queries `/api/v1/trust/query?name={name}&type=mcp_server` (200). MCP servers now show real trust scores from the registry. | Low | 3/15/2026 |
| v0.7.0 | ## Registry Enrichment & Shared Library Consolidation ### New Features - **`opena2a detect --registry`** -- Enriches detected MCP servers with community trust scores from the live OpenA2A Registry. Shows trust data inline (e.g., `stripe Trust: 50/100`). Opt-in, 5-second timeout, graceful degradation. ### Infrastructure - **Community scan-report endpoint live** -- `POST /api/v1/trust/scan-report` accepts scan results from all CLI tools. Rate limited (100 req/min), anonymous, no auth required. | Low | 3/15/2026 |
| v0.6.3 | ## Shadow AI Detection & Identity Fixes ### New Features - **`opena2a detect`** โ Shadow AI agent audit with governance scoring (0-100) - **`--report`** โ HTML executive report for enterprise audiences - **`--export-csv`** โ Asset inventory export for CMDB/ServiceNow with hostname, username, timestamp - **Shadow AI tab** in `opena2a review` (Phase 6 of 6) - 20+ AI agent/LLM detection patterns (Claude Code, Cursor, Copilot, Windsurf, Ollama, LM Studio, etc.) - Claude plugin MCP server discovery | Low | 3/15/2026 |
| v0.5.12 | ## Changes - Trust score now displays as percentage (e.g., `50%` instead of `0.5`) - Package type shows human-friendly labels (`MCP Server` instead of `mcp_server`) - Uses `displayType` from API when available, with local fallback mapping | Low | 3/14/2026 |
| v0.5.11 | ## opena2a-cli v0.5.11 ### Full aim-core Identity Coverage & Cross-Tool Integration **Identity commands** (12 subcommands): - `list`, `create`, `trust`, `audit`, `log`, `policy`, `check`, `sign`, `verify` - `attach` โ connect security tools to AIM identity with selective enablement (`--tools`) - `detach` โ disconnect tools and clear trust hints - `sync` โ refresh trust hints and import new tool events **Cross-tool integration:** - Event bridges: Shield, ARP, HMA, ConfigGuard, Secretless event | Low | 3/13/2026 |
| v0.5.8 | ## Fixed - `claim` command now defaults source to 'npm' before registry lookup, matching `trust` behavior. Previously `claim express` failed with "No trust profile found" while `trust express` succeeded. - `trust --verbose` now shows request URL, response time (ms), agent ID, source, and version. Previously `--verbose` produced identical output to the default view. | Low | 3/12/2026 |
| v0.5.7 | ## Changes since v0.5.6 - Add help text examples for trust and claim commands - Add GitHub URL auto-parsing (https://github.com/org/repo auto-detects --source github) - Add verbose output on error paths (registry URL, request path, full error) - List valid --source values in help text | Low | 3/12/2026 |
| v0.5.6 | ## Changes since v0.3.3 - Add `trust` command for agent trust profile lookup - Add `claim` command for agent ownership verification - Ed25519 keypair generation and local key storage - --json flag alias on trust/claim commands - Source validation, empty input handling - Improved error messages with actionable next steps - Demo GIF updates with JetBrains Mono font | Low | 3/12/2026 |
| v0.3.3 | ## What's New **`opena2a protect` is now a single command to fix all auto-fixable findings.** After `opena2a init` diagnoses your project, run `opena2a protect` to fix everything in one step. ### New: Unified Fix-Everything Flow ```bash opena2a init # Diagnose (read-only) opena2a protect # Fix everything fixable opena2a init # Re-assess -- watch your score improve ``` ### What `protect` now fixes - **Credentials** -- Detect, vault, and replace hardcoded secrets with env var refere | Low | 3/4/2026 |
| v0.3.1 | ## What's New **17 bug fixes** from a deep QA review covering security, correctness, and UX improvements across the CLI. ### Highlights - **Review dashboard redesign**: Score breakdown now uses structured explainers instead of generic stat cards - **Natural language input fix**: `opena2a find secrets` and `opena2a detect credentials` now work without shell quotes - **Project-scoped Shield events**: Events stored in `.opena2a/shield/` per-project instead of global `~/.opena2a/` - **Security fi | Low | 3/3/2026 |
| v0.1.2 | DRIFT-002: AWS access key liveness verification with Bedrock drift detection. Manual SigV4 signing, no AWS SDK. 325 tests, 18 new for DRIFT-002. | Low | 3/2/2026 |
