Athena misinterpreted generic per-finding option labels as execution agent names. Rewrite Step 3 "Mode: One by one" to use descriptive remediation approach labels derived from synthesis, add instruction block clarifying options vs execution methods, and show all three cases (multiple/single/no fix suggestions) in the example.
Replace the 3-route system (self-answerable/ambiguous/council-clear) with
6-category first-interaction routing (meta, wrong-agent, council-worthy,
simple/factual, tool/action, ambiguous) plus 2 pre-checks for explicit
opt-out and council requests. Add subsequent-interaction rules for direct
answers, auto-routing, and greetings. Add 'Cross-check with council'
action to all 8 guidance files for post-synthesis re-evaluation.
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
Add <agent_handoff> block to ATHENA_SYSTEM_PROMPT documenting capabilities
and constraints for Prometheus, Atlas, Sisyphus, and Hephaestus agents.
Prometheus entry explicitly states READ-ONLY constraint with forbidden
context verbs. Fix runtime guidance labels from 'Execute (Prometheus)'
to 'Plan (Prometheus)' with /start-work references.
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
The prompt showed only the opening <COUNCIL_MEMBER_RESPONSE> tag but
never the closing </COUNCIL_MEMBER_RESPONSE>, causing the test
assertion to fail. Added a concrete example block showing both tags.
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().
- Add _isCompleting guard and writeTaskOutput call before status flip
- Propagate writeOutputToFile from LaunchInput to BackgroundTask
- Delete council-result-format.ts and tests (no longer needed)
- 38/38 tests passing
- council_finalize: batch extract, archive, and return council results
- council_read: secure file reader for archived council responses
- background_wait: stripped to metadata-only, returns completed_tasks array
- Removed all council-specific logic from background_wait
- 28/28 tests passing across all changed files
ismeth