The v1 SDK Tui class has no selectSession method (only exists in v2).
Commit 4150e26d regressed this by replacing the working _client.post()
direct HTTP call with client.tui.selectSession() which silently failed.
Reverts to bypassing the SDK and POSTing to /tui/select-session directly,
which works regardless of SDK version.
Council members now skip TTL check in poller via isCouncilMemberAgent()
prefix match, covering both interactive (task tool) and bulk (athena_council)
launch paths. Users can cancel manually if needed.
Also fix promptFileMoved variable scoping: move declaration before try block
so finally block can access it. Dynamic timeout in error messages.
Add universal YAML front-matter header to all synthesis.md outputs (council,
question, date, members, session_ids, mode, intent, responded count).
AUDIT: structured finding format with required fields (severity, confidence,
members reported, evidence, impact, fix direction). New cross-check flow asks
user to verify minority/solo findings via a second council round before
proceeding to findings processing.
Also fix prompt-file cleanup: skip cleanupPromptFile when movePromptFile
already succeeded.
Add optional ttl field to BackgroundTask/LaunchInput. Council launcher
passes COUNCIL_MEMBER_TTL_MS (2 hours) per member. Task poller uses
task.ttl ?? TASK_TTL_MS, replacing the blanket TTL_EXEMPT_AGENTS bypass
for council members with a bounded timeout.
Co-authored-by: Vacbo <2445>
Remove blind 30-min setTimeout cleanup from prepare_council_prompt. Prompt
files now stay until council_finalize archives them in a try/finally block.
Extract shared resolvePromptTempFilePath helper used by both movePromptFile
and the new cleanupPromptFile safety net.
Co-authored-by: Vacbo <2445>
The background_wait instruction was only shown when the caller passed
run_in_background=false (forced case). But callers passing true also need
the guidance since they were polling background_output instead. Now
athena-junior always gets background_wait instructions regardless of how
it was launched.
When athena-junior is force-backgrounded (caller passed run_in_background=false),
the response now instructs the caller to use background_wait instead of
background_output, preventing repeated polling loops.
Council sessions are long-running: athena-junior launches members then
waits for them via background_wait. Two changes prevent premature kills:
- Force run_in_background=true when subagent_type resolves to athena-junior,
avoiding the 10-minute sync poll timeout (MAX_POLL_TIME_MS)
- Exempt athena-junior from the 30-minute TASK_TTL_MS hard cutoff since
council members have their own independent TTLs
- Fix missing bulk_launch field in AthenaOverrideConfigSchema (type error)
Athena primary should not be routable via switch_agent or appear in the
delegation table. Only athena-junior (subagent) remains delegatable via task().
- Remove athena from ALLOWED_AGENTS and DESCRIPTION in switch-agent tool
- Remove ATHENA_PROMPT_METADATA from agentMetadata, agent.ts, barrel export
- Update switch-agent test to assert athena rejection
- Remove ATHENA_PROMPT_METADATA test block
- Create src/tools/athena-council/ with council-launcher, tool factory, types
- Launch all council members in parallel via BackgroundManager with writeOutputToFile: true
- Wire tool into registry with backgroundManager + councilConfig dependencies
- Grant athena-junior athena_council permission, remove task permission
- Update non-interactive prompt to use athena_council instead of task tool
- Add explicit stop-after-result instruction to prevent post-output text
- 23 tests (14 tool + 9 launcher) using real temp files, no mock.module
Split the Athena council orchestrator into two distinct agents:
- athena: primary mode, interactive prompt with Question tool and switch_agent
- athena-junior: subagent mode, non-interactive prompt with structured JSON output
Key changes:
- Create athena-junior-agent.ts factory (mode=subagent, denies question+call_omo_agent)
- Revert athena agent.ts to primary mode (no env var switching)
- Make buildAthenaRuntimeGuidance mode-aware (strips action_paths for non-interactive)
- Add mode parameter to council_finalize tool
- Register athena-junior in builtin-agents, tool-config, model-requirements
- Replace "athena" with "athena-junior" in call_omo_agent ALLOWED_AGENTS
- Update all tests for new architecture
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Remove contextCollector dependency from council_finalize and append
Athena runtime guidance directly to the tool's JSON response string.
This fixes the timing bug where guidance registered via contextCollector
only arrived on the next message cycle, too late for Athena's same-turn
synthesis.
- Remove ContextCollector import, RegisterContext type, and collector.register() block
- Append buildAthenaRuntimeGuidance() output after JSON result with \n\n separator
- Simplify createCouncilFinalize() factory call in tool-registry (no options)
- Update unit tests: replace collector mock assertions with direct output string checks
- Update integration tests: wrap JSON.parse with extractJson() helper for hybrid output
council_read was a sandboxed readFile wrapper for .sisyphus/ archives.
Now that council_finalize extracts member responses into plain files,
the standard Read tool serves the same purpose. Removes the tool, its
tests, and updates Athena's prompt to reference Read directly.
Adopt the opencode-handoff pattern: switch_agent now creates a new session
and sends the context via promptAsync immediately in the tool's execute
function, then navigates the TUI to the new session.
Removes ~1700 LOC of complexity: persistent state management, event-based
hook monitoring, retry logic, session idle waiting, apply verification,
and terminal detection — all replaced by 3 SDK calls.
Deleted: src/features/agent-switch/ (14 files), src/hooks/agent-switch/ (4 files)
Updated: Athena prompt to announce handoff before calling switch_agent
Add DIAGNOSE, CREATE, PERSPECTIVES, FREEFORM intents to Athena's council
classification system. Refine existing AUDIT (add DIAGNOSE boundary),
EVALUATE (improve criteria guidance), and EXPLAIN (remove catch-all status).
Fix bug where prepare_council_prompt defaulted to AUDIT when no intent was
specified — this caused analytical frameworks to be injected into creative
requests (e.g. poetry). Now defaults to no addendum.
Derive validIntents from Object.keys(COUNCIL_INTENT_ADDENDUMS) to prevent
drift between the addendums record and the validation array.
task-output-writer wrote to process.cwd()/.sisyphus/task-outputs/ while
council_finalize read from ctx.directory/.sisyphus/task-outputs/. When
these differ, all output files appear missing.
Fix by passing directory to writeTaskOutput from BackgroundManager, and
adding basePath to council_read. Also removes process.chdir() from tests
that was masking the bug.
Ripgrep's --glob flag silently returns no results when the pattern
contains a directory prefix (e.g. 'apps/backend/**/*.ts'). Models
naturally write patterns this way when exploring unfamiliar codebases.
Extract the static directory segments from the pattern, append them to
the search path, and pass only the glob portion to ripgrep.
- Add council-continuation-enforcer to nudge council members until they
output COUNCIL_MEMBER_RESPONSE, replacing the hacky post-compaction hack
- Add background_wait + background_cancel to council member tool allowlist
so members can properly manage their explore agents before responding
- Update COUNCIL_DELEGATION_ADDENDUM prompt to instruct members to cancel
pending tasks and wait for explore results before final response
- Fix council_finalize path resolution to use project directory
- Remove post-compaction-continuation.ts and recentlyCompactedSessions hack
- Add council-response-checker for detecting response tag in session messages
When models pass relative paths (e.g. 'apps/ios/CleanSlate') to glob/grep
tools, they were passed directly to ripgrep which resolved them against
process.cwd(). In OpenCode Desktop, process.cwd() is '/' causing all
relative path lookups to fail with 'No such file or directory'.
Fix: use path.resolve(ctx.directory, args.path) to resolve relative paths
against the project directory instead of relying on process.cwd().
When models pass relative paths (e.g. 'apps/ios/CleanSlate') to glob/grep
tools, they were passed directly to ripgrep which resolved them against
process.cwd(). In OpenCode Desktop, process.cwd() is '/' causing all
relative path lookups to fail with 'No such file or directory'.
Fix: use path.resolve(ctx.directory, args.path) to resolve relative paths
against the project directory instead of relying on process.cwd().
ismeth