- Add worktree module for git worktree management
- Refactor do skill scripts from shell to Python for better maintainability
- Add install.py for do skill installation
- Update stop-hook to Python implementation
- Enhance executor with additional configuration options
- Update CLAUDE.md with first-principles thinking guidelines
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Add AllowedTools/DisallowedTools fields to AgentModelConfig and Config
- Update ResolveAgentConfig to return new fields
- Pass --allowedTools/--disallowedTools to claude CLI in buildClaudeArgs
- Add fields to TaskSpec and propagate through executor
- Fix backend selection when taskSpec.Backend is specified but backend=nil
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
ensureExecutableTempDir is intentionally no-op on Windows,
so tests should be skipped on that platform.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- CI: fetch tags for version detection
- Makefile: inject version via ldflags
- Add CODEAGENT_TMPDIR support for macOS permission issues
- Inject ANTHROPIC_BASE_URL/API_KEY for claude backend
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Replace echo with createFakeCodexScript() or fake command runner
- Use PID offsets based on os.Getpid() to avoid collisions in cleanup tests
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Use cmd.exe /c to execute .bat/.cmd on Windows
- Set USERPROFILE alongside HOME for os.UserHomeDir()
- Use setTempDirEnv to set TEMP/TMP on Windows
- Replace chmod-based tests with cross-platform alternatives
- Fix concurrent speedup benchmark with fair comparison
- Add output/ to gitignore
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Generate platform-specific mock executables in tests:
- Windows: codex.bat with @echo off
- Unix: codex.sh with #!/bin/bash
Fixes CI failures on windows-latest runner.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Add multi-platform testing (Ubuntu, Windows, macOS) to CI workflow.
Add unit tests for cross-platform path handling, stdin mode triggers,
and codex command construction to address issue #137.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
REMOVED all hardcoded default backend/model values from defaultModelsConfig.
Now ~/.codeagent/models.json is REQUIRED - missing config returns clear error
with example configuration.
BREAKING CHANGE: Users must configure ~/.codeagent/models.json before using
--agent or parallel tasks with agent: field.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
The pattern 'codeagent-wrapper' was matching cmd/codeagent-wrapper/
directory. Changed to '/codeagent-wrapper' to only match root binary.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Resolves intermittent "completed without agent_message output" errors
when Claude CLI exits before all stdout data is read.
- internal/executor/executor.go:43
- internal/app/app.go:27
- Add benchmark script for stability testing
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Change env var from ANTHROPIC_API_KEY to ANTHROPIC_AUTH_TOKEN
- Add Backend field propagation in taskSpec (cli.go)
- Add stderr logging for injected env vars with API key masking
- Add comprehensive tests for env injection flow
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
When using gemini backend without --model flag, now automatically
reads GEMINI_MODEL from ~/.gemini/.env file, consistent with how
claude backend reads model from settings.
Return exit code 1 when backend returns empty result.Message with exit_code=0.
Prevents silent failures where no output is produced.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
opencode does not support "-" as a stdin marker like codex/claude/gemini.
When using stdin mode, omit the "-" argument so opencode reads from stdin
without an unrecognized positional argument.
Closes#124
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Add codexNoisePatterns to filter "ERROR codex_core::codex: needs_follow_up:"
messages from stderr output when using the codex backend.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Replace invalid `--reasoning-effort` CLI flag with `-c model_reasoning_effort=<value>`
config override, as codex does not support the former.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Parallel task execution was not inheriting the --skip-permissions flag,
causing permission prompts to appear for parallel tasks while single
tasks worked correctly.
Changes:
- Add SkipPermissions field to TaskSpec struct
- Parse skip_permissions/skip-permissions in parallel task config
- Inherit SkipPermissions from CLI args to parallel tasks
- Pass SkipPermissions when creating task Config in executor
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Add 50ms sleep in createFakeCodexScript to ensure parser goroutine has
time to read stdout before the process exits. Fixes TestRun_ExplicitStdinSuccess
flaky failure on Linux CI where fast shell execution closes pipe prematurely.
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Add --reasoning-effort CLI flag for codex model thinking intensity
- Support reasoning config in ~/.codeagent/models.json per agent
- CLI flag takes precedence over config file
- Only effective for codex backend
Closes#117
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Claude: enable --dangerously-skip-permissions by default (set CODEAGENT_SKIP_PERMISSIONS=false to disable)
- Codex: enable --dangerously-bypass-approvals-and-sandbox by default (set CODEX_BYPASS_SANDBOX=false to disable)
- Gemini: use positional argument instead of deprecated -p flag (except for stdin mode)
- Add envFlagDefaultTrue helper for default-true env flags
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Add --agent parameter for agent-based backend/model resolution
- Add --prompt-file parameter for agent prompt injection
- Add opencode backend support with JSON output parsing
- Add yolo field in agent config for auto-enabling dangerous flags
- claude: --dangerously-skip-permissions
- codex: --dangerously-bypass-approvals-and-sandbox
- Add develop agent for code development tasks
- Add omo skill for multi-agent orchestration with Sisyphus coordinator
- Bump version to 5.5.0
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
- Add Model field to Config and TaskSpec for per-task model selection
- Parse --model flag and model: metadata in parallel tasks
- Auto-inject model from ~/.claude/settings.json for claude backend in new mode
- Pass --model to claude CLI, -m to gemini CLI, --model to codex CLI
- Preserve --setting-sources "" isolation while reading minimal safe subset
- Add comprehensive tests for model parsing, propagation, and settings injection
Fixes#105
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
This commit addresses the remaining Minor issues from PR #94 code review:
Minor #12: Unicode Symbol Compatibility
- Added CODEAGENT_ASCII_MODE environment variable support
- When set to "true", uses ASCII symbols: PASS/WARN/FAIL
- Default behavior (unset or "false"): Unicode symbols ✓/⚠️/✗
- Updated help text to document the environment variable
- Added tests for both ASCII and Unicode modes
Implementation:
- executor.go:514: New getStatusSymbols() function
- executor.go:531: Dynamic symbol selection in generateFinalOutputWithMode
- main.go:34: useASCIIMode variable declaration
- main.go:495: Environment variable documentation in help
- executor_concurrent_test.go:292: Tests for ASCII mode
- main_integration_test.go:89: Parser updated for both symbol formats
Minor #13: Performance Optimization - Reduce Repeated String Operations
- Optimized Message parsing to split only once per task result
- Added *FromLines() variants of all extractor functions
- Original extract*() functions now wrap *FromLines() for compatibility
- Reduces memory allocations and CPU usage in parallel execution
Implementation:
- utils.go:300: extractCoverageFromLines()
- utils.go:390: extractFilesChangedFromLines()
- utils.go:455: extractTestResultsFromLines()
- utils.go:551: extractKeyOutputFromLines()
- main.go:255: Single split with reuse: lines := strings.Split(...)
Backward Compatibility:
- All original extract*() functions preserved
- Tests updated to handle both symbol formats
- No breaking changes to public API
Test Results:
- All tests pass: go test ./... (40.164s)
- ASCII mode verified: PASS/WARN/FAIL symbols display correctly
- Unicode mode verified: ✓/⚠️/✗ symbols remain default
- Performance: Single split per Message instead of 4+
Usage Examples:
# Unicode mode (default)
./codeagent-wrapper --parallel < tasks.txt
# ASCII mode (for terminals without Unicode support)
CODEAGENT_ASCII_MODE=true ./codeagent-wrapper --parallel < tasks.txt
Benefits:
- Improved terminal compatibility across different environments
- Reduced memory allocations in parallel execution
- Better performance for large-scale parallel tasks
- User choice between Unicode aesthetics and ASCII compatibility
Related: #94
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Merging PR #94 with code review fixes applied.
All Critical and Major issues from code review have been addressed:
- 11/13 issues fixed (2 minor optimizations deferred)
- Test coverage: 88.4%
- All tests passing
- Security vulnerabilities patched
- Documentation updated
The code review fixes have been committed to pr-94 branch and are ready for integration.
- Fix incorrect filename reference from setting.json to settings.json in backend.go
- Update corresponding test fixtures to use correct filename
- Bump version from 5.2.7 to 5.2.8
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
Changes in v5.2.7:
- Security fix: pass env vars via process environment instead of command line
- Prevents ANTHROPIC_API_KEY leakage in ps/logs
- Add SetEnv() interface to commandRunner
- Type-safe env parsing with 1MB file size limit
- Comprehensive test coverage for loadMinimalEnvSettings()
Related: #89, PR #92
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
* fix: allow claude backend to read env from setting.json while preventing recursion
Fixes#89
Problem:
- --setting-sources "" prevents claude from reading ~/.claude/setting.json env
- Removing it causes infinite recursion via skills/commands/agents loading
Solution:
- Keep --setting-sources "" to block all config sources
- Add loadMinimalEnvSettings() to extract only env from setting.json
- Pass env explicitly via --settings parameter
- Update tests to validate dynamic --settings parameter
Benefits:
- Claude backend can access ANTHROPIC_API_KEY and other env vars
- Skills/commands/agents remain blocked, preventing recursion
- Graceful degradation if setting.json doesn't exist
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
* security: pass env via process environment instead of command line
Critical security fix for issue #89:
- Prevents ANTHROPIC_API_KEY leakage in process command line (ps)
- Prevents sensitive values from being logged in wrapper logs
Changes:
1. executor.go:
- Add SetEnv() method to commandRunner interface
- realCmd merges env with os.Environ() and sets to cmd.Env
- All test mocks implement SetEnv()
2. backend.go:
- Change loadMinimalEnvSettings() to return map[string]string
- Use os.UserHomeDir() instead of os.Getenv("HOME")
- Add 1MB file size limit check
- Only accept string values in env (reject non-strings)
- Remove --settings parameter (no longer in command line)
3. Tests:
- Add loadMinimalEnvSettings() unit tests
- Remove --settings validation (no longer in args)
- All test mocks implement SetEnv()
Security improvements:
- No sensitive values in argv (safe from ps/logs)
- Type-safe env parsing (string-only)
- File size limit prevents memory issues
- Graceful degradation if setting.json missing
Tests: All pass (30.912s)
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
---------
Co-authored-by: SWE-Agent.ai <noreply@swe-agent.ai>
- Add timestamp prefix format [YYYY-MM-DD HH:MM:SS.mmm] to every log entry
- Resolves issue where logs lacked time information, making it impossible to determine when events (like "Unknown event format" errors) occurred
- Update tests to handle new timestamp format by stripping prefixes during validation
- All 27+ tests pass with new format
Implementation:
- Modified logger.go:369-370 to inject timestamp before message
- Updated concurrent_stress_test.go to strip timestamps for format checks
Fixes#81
Generated with SWE-Agent.ai
Co-authored-by: SWE-Agent.ai <noreply@swe-agent.ai>
* Improve backend termination after message and extend timeout
* fix: prevent premature backend termination and revert timeout
Critical fixes for executor.go termination logic:
1. Add onComplete callback to prevent premature termination
- Parser now distinguishes between "any message" (onMessage) and
"terminal event" (onComplete)
- Codex: triggers onComplete on thread.completed
- Claude: triggers onComplete on type:"result"
- Gemini: triggers onComplete on type:"result" + terminal status
2. Fix executor to wait for completion events
- Replace messageSeen termination trigger with completeSeen
- Only start postMessageTerminateDelay after terminal event
- Prevents killing backend before final answer in multi-message scenarios
3. Fix terminated flag synchronization
- Only set terminated=true if terminateCommandFn actually succeeds
- Prevents "marked as terminated but not actually terminated" state
4. Simplify timer cleanup logic
- Unified non-blocking drain on messageTimer.C
- Remove dependency on messageTimerCh nil state
5. Revert defaultTimeout from 24h to 2h
- 24h (86400s) → 2h (7200s) to avoid operational risks
- 12× timeout increase could cause resource exhaustion
- Users needing longer tasks can use CODEX_TIMEOUT env var
All tests pass. Resolves early termination bug from code review.
Co-authored-by: Codeagent (Codex)
Generated with SWE-Agent.ai
Co-Authored-By: SWE-Agent.ai <noreply@swe-agent.ai>
---------
Co-authored-by: SWE-Agent.ai <noreply@swe-agent.ai>
