🤖 Native Sisyphus prompts for Claude Opus 4.7 and Kimi K2.x
If you run Sisyphus on Claude Opus 4.7 or Kimi K2.6, the prompt is now hand-tuned per model instead of reusing the generic Opus chain. Both variants tighten the delegation contract and reinforce the "stop guessing, use the tools" discipline that frontier models keep slipping on.
🛠️ Hephaestus GPT-5.5 rewrite
Hephaestus on GPT-5.5 has been re-architected around outcome-first delegation with a Codex 5.2 / Amp prompt distillation. End-to-end manual QA on full-delegation tasks is now mandatory — agents that own the deliverable also have to actually USE it before claiming done. Surface-matched QA tools (terminal for TUI, Playwright for web, curl for HTTP) are part of the contract, not optional.
⚙️ Quality-of-life
gpt-5.5is now the default fororacle,hephaestus, anddeepcategories (with auto-migration fromgpt-5.4).- Ralph loop no longer fires idle-continuation while background tasks are still running.
doctorparses multi-slash model IDs and stops warning on resolved provider capabilities.- Frontier agents (Opus 4.7, dotted Opus, GPT-5.5 Sisyphus) get the right tool whitelist; stale frontier tool denies are cleared on agent build.
posthogstartup no longer crashes on Bun runtimes whereos.cpus()hits a/sys/firmwarepermission error.- Several smaller fixes across tmux fallback ports, plugin handlers, file-reference env-var expansion, skill resolution after agent overrides, and dev-browser provider gating.
Minor Compatibility and Stability Release
This release carries compatibility-facing behavior changes and operational hardening. Read the summary below before upgrading or publishing.
- Rename transition updates across package detection, plugin/config compatibility, and install surfaces.
- Task and tool behavior updates, including delegate-task contract and runtime registration behavior.
- Task-system default behavior alignment so omitted configuration behaves consistently across runtime paths.
Commit Summary
- 4eb597b Merge pull request #3698 from code-yeongyu/fix/posthog-cost-reduction
- 90c2c9c Merge pull request #3697 from code-yeongyu/feat/sisyphus-kimi-k2-6
- 669e066 fix(telemetry): dedupe plugin_loaded event to once per UTC day
- 7a3a835 feat(sisyphus): add Kimi K2.x prompt variant
- 58e4b8f fix(telemetry): remove PostHog HAU tracking, keep DAU only
- 84287f1 test(todo-continuation): make countdown toast test deterministic
- 2483dc7 fix(agents): restore Opus Sisyphus prompt contracts
- c3fabaa refactor(agents): rewrite Hephaestus gpt-5.5 around outcome-first delegation contract
- 886845b @FurryWolfX has signed the CLA in #3695
- 6103df0 feat(agents): restore deep-work emphasis on Hephaestus gpt-5.5
- 7a89b98 refactor(agents): re-style Hephaestus gpt-5.5 in Codex 5.2 prompt structure
- 4aae402 refactor(agents): rewrite Hephaestus gpt-5.5 prose-dense from Codex 5.2 + Amp distillation
- 1365173 @javimarttinn has signed the CLA in #3687
- 6775dd1 feat(agents): mandate end-to-end usage validation on Hephaestus gpt-5.5 delegation
- c3b98c4 feat(agents): mandate surface-matched manual QA tools on Opus 4.7 full delegation
- 9e7b391 feat(agents): require manual end-to-end QA on full-delegation Opus 4.7 tasks
- 9feb9f6 refactor(agents): distill SMART production-prompt patterns into Opus 4.7 Sisyphus
- 241fb70 test(delegate-task): add background-task test coverage
- e1b2fa9 test(call-omo-agent): add sync-executor test coverage
- bda7fdb test(background-agent): add spawner test coverage
- 2024c0c feat(agents): wire Claude Opus 4.7 support into Sisyphus agent
- 41d4cbe feat(agents): add Claude Opus 4.7 native Sisyphus prompt builder
- 215c8f2 feat(agents): add isClaudeOpus47Model type guard
- 8ad50b7 docs(agents): replace ZWSP policy with sort-shim policy
- 333ad3a refactor(agents): drop ZWSP prefixes from agent display names
- dffe5a3 feat(agents): add narrowly-scoped agent sort shim and install at plugin entry
- 557d66f Merge pull request #3675 from code-yeongyu/fix/ralph-loop-remove-recovery-window
- c2a41a3 Merge pull request #3674 from code-yeongyu/fix/doctor-parse-multi-slash-model
- fe428ab Merge pull request #3673 from code-yeongyu/fix/single-task-directive-no-refuse
- bde2955 fix(ralph-loop): remove stale error recovery window
- 69c37e3 test(ralph-loop): cover non-abort error continuation
- e95a37f fix(doctor): parse multi-slash model IDs
- c27d52e fix(atlas): replace single task refusal directive
- 2e6b3b7 Merge pull request #3669 from code-yeongyu/fix/ralph-loop-zwsp-agent
- baf3e93 Merge pull request #3671 from code-yeongyu/fix/file-reference-env-var-expansion
- 1f9d9e3 Merge pull request #3672 from code-yeongyu/fix/ralph-loop-bg-task-guard
- ae7cdcc Merge pull request #3670 from code-yeongyu/fix/default-agent-config-key-normalization
- d0dee70 fix(ralph-loop): skip idle continuation during background tasks
- f429e53 test(ralph-loop): cover background task idle guard
- a46b7b8 fix(file-reference-resolver): expand env vars in path references
- 9081475 fix(plugin-handlers): normalize default agent runtime lookup
- c92f841 fix(ralph-loop): normalize continuation agent names
- 5e4aa8b Merge pull request #3667 from code-yeongyu/fix/dev-browser-provider-gating
- f1d4eb7 fix(skill-context): filter dev-browser as provider-gated
- 0d3eed1 fix(builtin-skills): gate dev-browser by provider selection
- a835fef Merge pull request #3665 from code-yeongyu/fix/posthog-arch-cpus-crash
- ee260fb Merge pull request #3666 from code-yeongyu/fix/minimax-kimi-thinking-param
- 194d668 Merge pull request #3668 from code-yeongyu/fix/skills-override-resolution
- dbf0bb9 test(agents): cover skills override prompt injection
- aeb4419 fix(agents): resolve skills after agent overrides
- 71c29c8 fix(model-capabilities): disable thinking for minimax and non-thinking kimi
- d1bc25a fix(posthog): guard CPU telemetry collection
- d32300a @islee23520 has signed the CLA in #3664
- 1da7df1 Revert "Merge pull request #3657 from code-yeongyu/refactor/replace-zwsp-with-real-spaces"
- 9823462 Merge pull request #3661 from code-yeongyu/fix/doctor-heuristic-model-status
- a696c8a fix(hooks): apply Sisyphus GPT-5.5 variant
- f1a11f2 Merge pull request #3657 from code-yeongyu/refactor/replace-zwsp-with-real-spaces
- 136a54f fix(doctor): avoid warnings for resolved model capabilities
- 8e78596 fix(model-capabilities): recognize current provider aliases
- f100a85 fix(agents): keep object keys clean by separating list display from runtime name
- 62c19ce Merge pull request #3659 from code-yeongyu/fix/dotted-opus-frontier-tools
- 80791f1 fix(agents): cover dotted opus frontier model
- 3f30dac Merge pull request #3656 from code-yeongyu/feature/hide-grep-glob-for-frontier-agents
- acf293d fix(tmux): validate fallback port
- ac8987a fix(agents): preserve explicit legacy tool denies
- baef011 @LathissKhumar has signed the CLA in #3658
- c65f90e fix(tmux): preserve configured fallback port
- 4672485 fix(agents): clear stale frontier tool denies
- 0e9bb59 refactor(agents): replace broken ZWSP sort prefixes with leading ASCII spaces
- 8a49a03 test(skill): fresh import skill tool factory
- 1680578 test(skill): stabilize command discovery spy
- a2c3804 test(slashcommand): isolate command discovery mocks
- ecce65b fix(skill): isolate injected skill discovery
- f1353cd fix(tmux): use stable fallback server URL
- c46b712 fix(agents): hide grep glob for frontier agents
- f74d03c Merge pull request #3654 from acamq/fix/failing-ci
- 25e1fa9 @gutierrezx7 has signed the CLA in #3651
- b32620c Merge pull request #3640 from ismetanin/docs/fix-kimi-code-link
- 6a0272f @ismetanin has signed the CLA in #3640
- 3577f30 docs(readme): fix Kimi Code subscription link and price
- c3b3417 @ihoooohi has signed the CLA in #3637
- 48b0cfe @darianstlex has signed the CLA in #3626
- 33ac355 Preserve migration history during config migration
- fbd4cfb Update OpenAI defaults to GPT-5.5
- 17d3350 @hackerh3 has signed the CLA in #3600
- 7bbe11a Merge pull request #3623 from MoerAI/fix/session-model-recovery-guard
- 7fa21ef fix(chat-message): remove unreachable guard that blocked session model recovery (fixes #3561)
- 6b46b9b fix(model-capabilities): add gpt-5.5 capability snapshot
- 3875b40 feat(migration): auto-upgrade openai/gpt-5.4 to openai/gpt-5.5
- 3bab66b feat(models): promote gpt-5.5 as default for oracle, hephaestus, and deep
- 1ff60d5 feat(agents): add gpt-5.5 native oracle prompt
- c57d08c feat(agents): add gpt-5.5 native sisyphus-junior prompt
- a432e29 feat(agents): add gpt-5.5 native hephaestus prompt
- 98964eb feat(agents): add gpt-5.5 native sisyphus prompt
- 69e6f38 feat(agents): reintroduce isGpt5_5Model for per-version prompt routing
- 81c5ed4 docs(drafts): add gpt-5.5 native prompt drafts
Thank you to 3 community contributors: