Replace oracle as the agent for council background tasks with a purpose-built
council-member agent. This avoids coupling to oracle's config/prompt and provides
proper read-only tool restrictions (deny write, edit, task, athena_council).
- New council-member-agent.ts with analysis-oriented system prompt
- Registered in agentSources (hidden from Sisyphus delegation table)
- Added to type system, Zod schemas, display names, tool restrictions
- Minimal model fallback (always overridden per council member at launch)
- Council orchestrator now launches members as council-member agent
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Athena was the only agent setting name explicitly. The mismatch between
the name property ('Athena (Council Orchestrator)') and the config key
('Athena (Council)') caused TypeError during agent resolution.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
LaunchInput.temperature and LaunchInput.permission were accepted and
passed through the council orchestrator but never forwarded to the
actual promptAsync API call (SDK doesn't support per-request temperature
or permission). Remove the dead fields, the unused AthenaConfig
interface, and update tests/docs/schema accordingly.
Already denied via agent-tool-restrictions.ts for all athena sessions,
but now also explicitly denied in the per-launch permission to make
the anti-recursion intent clear at the launch site.
Remove 9 files (913 lines) from the code-driven synthesis pipeline that
was superseded by the agent-driven approach in phases 6-8.
Phases 3/5 built: collectCouncilResults → formatForSynthesis →
buildSynthesisPrompt → formatFindingsForUser → buildDelegationPrompt.
Phases 6-8 replaced with: launch → background_output → Athena
synthesizes in conversation → switch_agent. The old pipeline was
never wired into runtime and all consumers were other dead code.
Also simplifies executeCouncil to return CouncilLaunchResult (task IDs
+ failures) instead of reading stale task status via collectCouncilResults.
Deleted: council-result-collector, synthesis-types, synthesis-prompt,
synthesis-formatter, findings-presenter, delegation-prompts (+ 4 tests).
Cleaned: CouncilMemberStatus, AgreementLevel, CouncilMemberResponse,
CouncilExecutionResult types from types.ts.
- Forward temperature and permission through council-launcher to background manager
- Add LaunchInput.temperature and LaunchInput.permission to background-agent types
- Extract session guard with 5-minute timeout to prevent stale council locks
- Make council optional in AthenaOverrideConfigSchema for partial user overrides
- Support member lookup by both name and model ID in filterCouncilMembers
- Add provider/model-id format validation to CouncilMemberSchema
- Fix findings-presenter group header to show finding count instead of first finding's reporter count
After Athena synthesizes council findings, presents user with Question tool
TUI to choose: Atlas (fix now), Prometheus (create plan), or no action.
On selection, session_handoff tool stores intent + calls updateSessionAgent(),
then agent-handoff hook fires on session.idle to switch the main session's
active agent via promptAsync with synthesis context.
- update Athena workflow to launch council then call background_output per task
- require collecting all member responses before synthesis and delegation
Prometheus needs to interview the user interactively, so it can't run as a
background task. Updated Athena's delegation prompt:
- Atlas: still delegates via task tool (autonomous execution)
- Prometheus: outputs structured findings summary and tells the user to
switch to Prometheus agent, which sees the conversation context and
can ask clarifying questions directly
The old prompt said 'when requiring multi-model analysis' which let Athena
decide to skip the council and do direct analysis herself. Combined with
keyword-detector injecting [search-mode] telling her to 'launch explore
agents and use Grep directly', Athena never called athena_council.
Two fixes:
1. System prompt now unconditionally requires athena_council as FIRST action
- Explicitly prohibits Read/Grep/Glob/LSP/call_omo_agent
- Identity is 'orchestrator, not analyst'
2. keyword-detector skips ALL injections for Athena agent
- search/analyze/ultrawork modes conflict with council orchestration
- Same pattern as isPlannerAgent() skip for Prometheus
- Build pure prompt constructors with confirmed finding context and agreement levels
- Add BDD tests for fix/planning intent, question context, and single-finding edge cases
- implement createAthenaAgent with primary-mode model behavior and prompt metadata
- export Athena factory and metadata through athena and root agent barrels
- Add synthesis result contracts with agreement, provenance, and Athena assessment fields\n- Add synthesis prompt builder and council-response formatter with failure-aware provenance output
- Cover completed, partial failure, total failure, and custom member naming scenarios\n- Assert provenance fields and response/error rendering requirements
- Implement executeCouncil with parallel member launch and partial-failure tolerance
- Add result collection mapping and wire Athena exports with read-only athena tool restrictions
- Add BDD coverage for parallel launch, partial failures, and invalid model handling
- Verify shared council prompt/model parsing inputs and per-member passthrough fields
- Add council execution result and member response types for orchestration
- Implement provider/model parser for BackgroundManager-compatible model input
- Add shared council prompt builder and export new athena modules
- Cover standard provider/model strings for supported council members
- Validate edge case handling for model IDs with extra slashes
- Assert null output for malformed parser inputs
- add AthenaOverrideConfigSchema so athena supports council plus standard override fields
- export athena schema/contracts and add root config tests for valid and invalid athena overrides
- switch schema generation to zod v4 toJSONSchema and regenerate JSON schema with athena council structure
- add athena to built-in and overridable agent name schemas
- extend BuiltinAgentName with athena for config-level recognition
- make builtin agent source maps partial until athena runtime registration lands
- Add Athena council config interfaces and execution status types
- Add standalone Zod schemas for council member, council, and top-level Athena config
- Enforce 2-member minimum and bounded optional temperature validation
The agent prompt described PDF handling but did not tell the agent
to call the Read tool, which is its only allowed tool. Added
explicit instruction so PDFs and documents are actually loaded
before extraction.
🤖 Generated with OhMyOpenCode assistance
https://github.com/code-yeongyu/oh-my-opencode
Anthropic API accepts claude-opus-4.6 (dot format) but not
claude-opus-4-6 (dash format). Added anthropic case to
transformModelForProvider to normalize dash-format model IDs before
they reach the provider. Updated model config snapshots and test
expectations to match the new dot-format output.
🤖 Generated with OhMyOpenCode assistance
https://github.com/code-yeongyu/oh-my-opencode
Extract hardcoded GPT apply_patch permission logic into a reusable module
to ensure consistent behavior across all agents. This prevents GPT models
from using the unreliable apply_patch tool while allowing other models.
- Add gpt-apply-patch-guard.ts with GPT_APPLY_PATCH_GUIDANCE and getGptApplyPatchPermission
- Update Hephaestus agent to use centralized permission logic
- Update Sisyphus-Junior agent to use centralized permission logic
- Update all GPT prompt builders to reference shared guidance constant
🤖 Generated with assistance of OhMyOpenCode
Built-in agents (Sisyphus, Hephaestus, Atlas) now include explicit
identity section in their system prompts, matching custom agent behavior.
12 agent identity tests pass, 266 total pass, tsc clean.
Closes#2964
GPT models (5.3-codex, 5.4, etc.) frequently hang when using apply_patch
due to verification loops. This adds:
1. Tool restriction: apply_patch is denied for GPT variants of
Hephaestus, Sisyphus-Junior, and Sisyphus agents
2. Prompt guidance: GPT-specific prompts now explicitly instruct using
edit/write tools instead of apply_patch
3. Removed the 'Always use apply_patch' instruction from
sisyphus-junior/gpt-5-4.ts that contradicted the fix
The deny is model-conditional — Claude variants retain apply_patch
access since it works reliably there.
Tool return values contained CTAs ("Use background_output to check") that
caused models to immediately poll instead of waiting for <system-reminder>
notifications. Unified all 5 tool return messages with explicit "Do NOT call
background_output now" guard, added pre-notification warning to BackgroundOutput
tool description, and strengthened Background Result Collection sections across
all 3 Sisyphus prompt variants (default, gpt-5-4, main) with BLOCKING
anti-pattern enforcement.
🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode)
ismeth