Agentic AI Atlasby a5c.ai
OverviewWikiGraphFor AgentsEdgesSearchWorkspace
/
GitHubDocsDiscord
iiRecord
Agentic AI Atlas · Daily Update - 2026-05-29
page:docs-daily-updates-2026-05-29a5c.ai
Search record views/
Record · tabs

Available views

II.Record viewspp. 1 - 1
overviewarticlejsongraph
III.Related pagespp. 1 - 1
II.
Page reference

page:docs-daily-updates-2026-05-29

Reading · 6 min

Daily Update - 2026-05-29 reference

Reference article for page:docs-daily-updates-2026-05-29 and its linked atlas edges.

Pagewiki/docs/daily-updates/2026-05-29.mdOutgoing · 0Incoming · 1

Daily Update - 2026-05-29

Summary of Changes

New Features

  • push all 10 adversarial dimensions to 100 (8cf40df99)
  • push all 10 adversarial dimensions to 95+ (f44398767)
  • server cache, barrel index, shared helper extraction (604adf0db)
  • 30 new tests, remove mock data, dynamic tool discovery (622b8060e)
  • fetch dedup, confirm dialog, unsaved changes, component docs (83343e4f5)
  • complete accessibility — 40→3 components with zero a11y (247665f4b)
  • router.refresh() migration, API route tests, force-dynamic fixes (fff3bfd54)
  • add meaningful accessibility attributes to 20 components (3031a8c4a)
  • add pagination to all list endpoints and UI (8beaf2162)
  • add evil fallback audit and here be dragons audit processes (1332cb106)
  • add route-level loading skeletons for 8 sections (f2a23c7d0)
  • add Getting Started onboarding page (332411dd4)
  • add transcript viewer to dispatch run detail page (7e5c0dafa)
  • rewrite external provider wizard and list for typed providers (3e783b005)
  • replace ExternalBackendProvider with 5 domain-scoped provider kinds (af3843f11)
  • add inference playground with side-by-side model comparison (58474fad1)

Bug Fixes

  • harden remaining critical+high evil fallbacks across repo (ddf00cfb8)
  • route opencode through proxy for non-bundled providers (184bda9b0)
  • cap CLI orchestration at 20 iterations, bail on no-effects loop (ebd83c458)
  • remove mini model silent substitution, log non-zero exit tolerance (eb67d3984)
  • paginate agent runs/sessions to 25 items, aria on issue editor (722d2073c)
  • write opencode config as file, not just env var (46cdb2450)
  • fix real audit issues — 5 unauthed GETs, 8 confirm(), 5 silent catches (dc3e3e4bb)
  • add GH_TOKEN, CURSOR_API_KEY, COPILOT_GITHUB_TOKEN to live-stack workflow (a9c29d5cb)
  • single-task process template and 120s iterate timeout (c31b6381d)
  • remove silent matrix defaults, assertion weakening, and config cascades (bace9627e)
  • add metadata to 3 issue pages, error boundaries to 10 route groups (8f40b2dbb)
  • remove 5 evil fallbacks that hide real problems (90df526cf)
  • add diagnostic logging to 49 critical+high evil fallbacks across all packages (5758e11df)
  • let hermes call Google directly, mark foundry/anthropic as blocked (4bbaed8cf)
  • use OPENROUTER_API_KEY and provider: openrouter for hermes (e33d6e123)
  • write hermes cli-config.yaml with proxy base_url (d8664b0e5)
  • add hermes provider translations for foundry/anthropic/google (2d62fab2e)
  • remove hermes custom provider config, use standard OpenAI env (0586b9b16)
  • disable proxy auth for hermes (like gemini-cli) (a5cd4a554)
  • log Gitea tree fallback instead of silently swallowing (70a0d6a55)
  • replace staging Atlas URL with production default (d31c2dc44)
  • replace silent .catch(() => {}) with console.warn logging (e9153da01)
  • revert hermes launch to stdin delivery with keep-open (8de323dd6)
  • adversarial audit fixes — security, errors, docs, accessibility (29c0dd327)
  • pass thoughtSignatureStore to /v1/responses and WebSocket paths (c14c4500d)
  • inject prompt flag in bridge-interactive for cli-flag harnesses (ece07b15f)
  • write hermes proxy config to ~/.hermes/ instead of temp dir (b550e3a54)
  • add auth to policy-reports, cache invalidation imports to profile (884f0c093)
  • batch 3 color migration — 465 total replaced (1058→593) (d0a46db4d)
  • batch 2 color migration + add 'use client' to inference-helpers (c37b7fffe)
  • hermes back to -z headless mode — stdin TUI doesn't execute tasks (284871025)
  • harden hook sandbox, add error boundaries, 6 adversarial tests (e6f08431c)
  • replace 231 hardcoded hex colors with CSS variables (c02c0393c)
  • write hermes proxy config to cli-config.yaml not config.yaml — fixes #468 (0b2aafe0d)
  • use positional defineTask form in raw-session template (c29a7e82f)
  • auto-correct babysitter-sdk import to @a5c-ai/babysitter-sdk (3ec8d665c)
  • resolve agent prompt from both string and object formats (82c38ac84)
  • post raw text as CLI orchestration task result (230eaf3af)
  • instruct raw agent-core process to use fs.writeFile for file output (71300ad56)
  • pin gemini-cli to 0.43.0 — 0.44.x auth regression confirmed (118589cca)
  • add CJS require fallback to orchestrateIteration process loader (8d90a7178)
  • fallback to CJS require when ESM import fails for process modules (52d6f49c5)
  • ensure .a5c/processes has package.json type:module, use execFileSync (b0dbc67c2)
  • pin gemini-cli to 0.43.0 + revert proxy env — fixes #483 (fbea9028e)
  • hermes stdinBehavior keep-open + needsIdleKill — fixes #468 (8e29dd91f)
  • hermes prompt via stdin instead of -z flag — fixes #468 (25cb23739)
  • use execFileSync for CLI orchestration to avoid shell escaping (2e5381068)
  • move hermes --output-format jsonl to unconditional block — fixes #468 (a2f7a5a9f)

Refactors

  • split settings-providers (425→259) and inference-playground (528→319) (f315b7e64)
  • split kanban-enhanced (602→264) and runner-pool-manager (434→241) (07e894410)
  • split assistant-chat.jsx (656→225 lines) into 3 modules (d1c2495bd)
  • split workspace-panel.jsx (704→219 lines) into 3 modules (765e828b8)
  • split artifact-registry.jsx (753→112 lines) into 4 modules (c0d2bdc85)
  • split 1254-line agent-pages.jsx into 11 focused modules (c970bcf16)
  • extract shared phaseTone helper, remove 7 duplicates (46538b84f)
  • split inference-service-manager into 7 focused files (2d86aae98)

Documentation

  • mark fixed evil fallbacks in evil-fallbacks.md (69e00ef14)

Maintenance

  • add orchestration trace logging for genty CI diagnosis (d1bbc3916)
  • regenerate dist-types and apply linter auto-fixes (4a9c8597c)
  • track upstream agent versions (#545) (2f9063d8d)

PRs Merged

  • #545 Track upstream agent CLI versions - https://github.com/a5c-ai/babysitter/pull/545

Issues Opened

  • #563 Live-stack: BP mode gemini-3.5-flash model behavior — agent runs but doesn't create output file (OPEN) - https://github.com/a5c-ai/babysitter/issues/563
  • #562 Live-stack: cursor-cli CI integration — needs investigation (OPEN) - https://github.com/a5c-ai/babysitter/issues/562
  • #561 Live-stack: opencode fails — server startup error in CI (OPEN) - https://github.com/a5c-ai/babysitter/issues/561
  • #560 Live-stack: copilot-cli fails — no GitHub auth configured in CI (OPEN) - https://github.com/a5c-ai/babysitter/issues/560
  • #547 Live-stack: gemini-cli BP and BI modes fail — prompt not injected in PTY bridge path (CLOSED) - https://github.com/a5c-ai/babysitter/issues/547
  • #546 Live-stack: Google API thought_signature missing in /v1/responses and WebSocket proxy paths (CLOSED) - https://github.com/a5c-ai/babysitter/issues/546
  • #544 Track upstream amp-0.0.1780032159-g3925ab release (OPEN) - https://github.com/a5c-ai/babysitter/issues/544
  • #543 Track upstream droid-0.136.1 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/543
  • #542 Track upstream omp-15.5.11 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/542
  • #541 Track upstream pi-0.77.0 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/541
  • #540 Track upstream opencode-1.15.12 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/540
  • #539 Track upstream gemini-cli-0.44.1 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/539
  • #538 Track upstream codex-0.135.0 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/538
  • #537 Track upstream claude-agent-sdk-0.3.156 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/537
  • #536 Track upstream claude-code-2.1.156 release (OPEN) - https://github.com/a5c-ai/babysitter/issues/536
  • #535 Track Amazon Nova 2 Pro model version (OPEN) - https://github.com/a5c-ai/babysitter/issues/535
  • #534 Track Gemini image model records and Gemini 2.x lifecycle deprecations (OPEN) - https://github.com/a5c-ai/babysitter/issues/534
  • #533 Track OpenAI GPT-5.5 Instant and chat-latest lifecycle updates (OPEN) - https://github.com/a5c-ai/babysitter/issues/533
  • #532 Track Claude Opus 4.8 model version (OPEN) - https://github.com/a5c-ai/babysitter/issues/532
  • #531 claude plugin update is a silent no-op — plugin version in marketplace.json never bumps across releases (OPEN) - https://github.com/a5c-ai/babysitter/issues/531

Issues Closed

  • #547 Live-stack: gemini-cli BP and BI modes fail — prompt not injected in PTY bridge path - https://github.com/a5c-ai/babysitter/issues/547
  • #546 Live-stack: Google API thought_signature missing in /v1/responses and WebSocket proxy paths - https://github.com/a5c-ai/babysitter/issues/546
  • #486 Live-stack: gemini-cli BI fails on Ubuntu/Windows (PTY fallback issue) - https://github.com/a5c-ai/babysitter/issues/486
  • #483 Live-stack: gemini-cli NI regression — "Invalid auth method selected" since May 28 - https://github.com/a5c-ai/babysitter/issues/483

CI Status Summary

  • Publish: pending - docs: update here-be-dragons.md with fix status (https://github.com/a5c-ai/babysitter/actions/runs/26657175843)
  • Publish: completed / cancelled - refactor(kradle-web): split inference-service-manager (473→224) with R… (https://github.com/a5c-ai/babysitter/actions/runs/26657147290)
  • Live Stack: in_progress - Live Stack (https://github.com/a5c-ai/babysitter/actions/runs/26657082415)
  • Publish: in_progress - docs: mark all remaining critical+high evil fallbacks as fixed (https://github.com/a5c-ai/babysitter/actions/runs/26657024915)
  • QA Daily: in_progress - fix: harden remaining critical+high evil fallbacks across repo (https://github.com/a5c-ai/babysitter/actions/runs/26657002229)
  • Publish: completed / cancelled - fix: harden remaining critical+high evil fallbacks across repo (https://github.com/a5c-ai/babysitter/actions/runs/26657002165)
  • Publish: completed / cancelled - refactor(kradle-web): split settings-providers (425→259) and inference… (https://github.com/a5c-ai/babysitter/actions/runs/26656977591)
  • Live Stack: completed / failure - Live Stack (https://github.com/a5c-ai/babysitter/actions/runs/26656792817)
  • Publish: completed / failure - fix(adapters): route opencode through proxy for non-bundled providers (https://github.com/a5c-ai/babysitter/actions/runs/26656783686)
  • Publish: completed / cancelled - refactor(kradle-web): split kanban-enhanced (602→264) and runner-pool-… (https://github.com/a5c-ai/babysitter/actions/runs/26656767865)

Live-Stack Test Results

  • ./.github/workflows/live-stack-published.yml
  • ./.github/workflows/live-stack.yml
  • ./scripts/live-stack-coverage-report.cjs

Article source

The article body is owned directly by this record.

Related pages

No related wiki pages for this record.

Shortcuts

Open overview
Open JSON
Open graph