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
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
- 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
Extract sendPostCompactionContinuation to dedicated file — council members now resume after compaction instead of silently failing. Refresh lastUpdate before async validation in both idle handler and polling path to prevent stale timeout from racing with completion detection.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
- Add session.compacted handler in BackgroundManager to prevent premature
task completion after compaction (defer first post-compaction idle)
- Explicitly block TodoWrite/TodoRead for council members in all sync
points (AgentConfig permission + session tools + prompt instructions)
- Add council member prefix check to todo-continuation-enforcer skip list
to prevent infinite continuation loops on completed council members
- Add optional analysis mode (solo/delegation) question to Athena setup:
solo = thorough but heavier, delegation = fast via explore/librarian
- Allow call_omo_agent in council member allow-list for delegation mode
- Update COUNCIL_MEMBER_PROMPT with TodoWrite prohibition and delegation
addendum for when delegation mode is selected
- Update prepare_council_prompt tool with mode parameter
- Make directory required in SkillLoadOptions, getAllSkills, and async
skill template resolvers to prevent unsafe process.cwd() fallback
- Remove dead skill export and process.cwd() fallback in skill tool
- Replace kimi-for-coding/k2p5 with kimi-for-coding/kimi-k2.5 in
council-members-generator
Skill discovery in the task tool failed to find project-level skills in
OpenCode Desktop because:
1. resolveSkillContent() never passed directory to the skill resolution
functions, causing them to fall back to process.cwd() which differs
from the project directory in Desktop mode.
2. getAllSkills() cache key only included browserProvider, not directory.
A first call with the wrong directory would cache stale results that
all subsequent calls (even with correct directory) would return.
3. The error message used discoverSkills() (discovered only) instead of
getAllSkills() (discovered + builtins), hiding builtin skills from
the Available list.
Changes:
- skill-resolver.ts: accept and pass directory; use getAllSkills for error msg
- tools.ts: pass options.directory to resolveSkillContent
- skill-discovery.ts: include directory in cache key; rename cache variable
- skill/types.ts + tools.ts: add directory to SkillLoadOptions for consistency
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.
- 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.
The opencode-config-agents-reader.test.ts was using mock.module() which
permanently replaced the module in bun's module cache, causing state
pollution in downstream tests (plugin-detection, write-omo-config,
config-loader). Replaced with spyOn() pattern that properly restores
in afterEach.
Blocking fixes:
- B1: Return empty restrictions for unknown/custom agents instead of
EXPLORATION_AGENT_DENYLIST, allowing custom agents full tool access
- B2: Use Object.create(null) consistently across all 5 agent-loading
result objects to prevent prototype pollution
- B3: Add code comment documenting custom agent bash access trust model
- B4: Mock getOpenCodeConfigDir in opencode-config-agents-reader tests
to prevent global config dir leakage
Non-blocking fixes:
- N1: Use resolveAgentDefinitionPaths with project boundary enforcement
in opencode-config-agents-reader for path containment
- N2: Add session-scoped 30s TTL cache to resolveCallableAgents to
avoid redundant SDK IPC calls per tool invocation
- N3: Extract shared parseToolsConfig into src/shared/parse-tools-config.ts
replacing 4 duplicated local implementations
- N4: Add .min(1) to AgentDefinitionPathSchema rejecting empty paths
- N5: Add resolve-agent-definition-paths.test.ts covering tilde expansion,
relative paths, boundary enforcement, and null containmentDir
- N6: Validate agent mode against allowed values instead of bare type
assertion in opencode-config-agents-reader
Addresses cubic-dev-ai P1 review: Object.hasOwn() alone doesn't prevent
a crafted '__proto__' agent name from mutating the result object's
prototype chain. Using Object.create(null) eliminates inherited properties
entirely, making both the hasOwn checks and property assignments safe.
Addresses cubic-dev-ai review: using 'in' on plain objects can skip valid
agent names that match inherited properties (toString, constructor, etc.).
Switched both occurrences in opencode-config-agents-reader.ts to
Object.hasOwn() for safe own-property checks.
- Case-insensitive .md extension stripping for agent name extraction
- Resolve project agent_definitions paths relative to config dir (.opencode/)
- Use getOpenCodeConfigDir() to respect OPENCODE_CONFIG_DIR/XDG_CONFIG_HOME
- First-write-wins semantics for both inline and definition-file agents
so project-level agents always take precedence over global-level
- Modified agent-config-handler.ts to load and integrate both new agent sources
- Added loadAgentDefinitions() and readOpencodeConfigAgents() calls in loading phase
- Integrated both sources into agent precedence chains (both Sisyphus-enabled and disabled paths)
- Added detailed logging for new agent sources
- Added filtering logic to respect disabled_agents configuration
- Extended agent-config-handler.test.ts with 7 new integration tests
- All tests passing (18/18 integration, 65/65 loader suite)
Wave 3 of agent definitions enhancement complete.
Wave 1 of agent definitions enhancement (PR #2299):
Schema & Configuration:
- Add agent_definitions field to oh-my-opencode config schema
- Support list of file paths to .md or .json agent definition files
- Add to PARTIAL_STRING_ARRAY_KEYS for Set-union merge semantics
- Implement eager path resolution in loadPluginConfig() before merging
Path Resolution:
- Create resolve-agent-definition-paths.ts helper
- User-level paths resolve from ~/.config/opencode/ (no containment)
- Project-level paths resolve from project root (with containment check)
- Homedir expansion, absolute/relative path handling
JSON Agent Loader:
- Create parseJsonAgentFile() for .json/.jsonc agent definitions
- Validate required fields (name, prompt)
- Support tools as string (comma-separated) or array
- Map model via mapClaudeModelToOpenCode()
- Comprehensive test suite (7 test cases, all passing)
Type Extensions:
- Extend AgentScope: add 'definition-file' and 'opencode-config'
- Add AgentJsonDefinition interface for JSON agent schema
All automated checks passing:
- lsp_diagnostics clean on all changed files
- json-agent-loader.test.ts: 7/7 passing
- Full typecheck: zero new errors
- QA evidence saved to .sisyphus/evidence/
- #3354: Coerce data.name to String in loadSkillFromPath/loadSkillFromPathAsync
to prevent crash when YAML parses numeric skill names (e.g., name: 12306)
- #3416: Add required run_in_background parameter to all task() examples in
ultrawork prompts (default, gpt, gemini, planner) to match tool schema
- #3379/#3417/#3418/#3337/#3335: Strip ZWSP (U+200B) before agent name
comparisons in agent-tool-restrictions, sync-prompt-sender, tool-execute-after,
tool-execute-before, oracle-verification-detector, call-omo-agent,
recovery-prompt-config, and agent-variant to prevent ZWSP-prefixed display
names from breaking exact-match lookups
- shell-env: detect Git Bash via MSYSTEM env var when SHELL is unset (#3366)
On some Git Bash installations SHELL is not set but MSYSTEM (MINGW64/MSYS)
is always present. Check MSYSTEM before PSModulePath to avoid emitting
PowerShell syntax in bash shells.
- session-state: resolve legacy agent names in resolveRegisteredAgentName (#3272)
Historical sessions stored agent names like 'Sisyphus (Ultraworker)' which
don't match the current registered format. Fall back to getAgentConfigKey
for legacy/parenthesized name resolution before returning the raw name.
- config-migration: skip backup when file content is unchanged (#3222)
Compare serialized config with existing file content before creating a
timestamped .bak file. Only create backup when the on-disk content
actually differs from the migrated content.
resolveSubagentSpawnContext and related lookups called client.session.get
without the query.directory parameter, causing project-scoped sessions
to 404 under newer OpenCode SDK versions. Threaded directory through
SpawnerContext/BackgroundManager so all four call-sites pass it.
🤖 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
ismeth