Changes ImageAnalysisConfig from providers array to provider_models
mapping for granular vision model control per CLIProxy provider.
Breaking change: config.yaml image_analysis section now uses
provider_models instead of providers/model fields.
Provider-to-model mappings:
- agy → gemini-2.5-flash
- gemini → gemini-2.5-flash
- codex → gpt-5.1-codex-mini
- kiro → kiro-claude-haiku-4-5
- ghcp → claude-haiku-4.5
- claude → claude-haiku-4-5-20251001
Hook checks CCS_CURRENT_PROVIDER against provider_models and skips
if no vision model configured for that provider.
Intercept Read tool calls for image/PDF files and route through CLIProxy
with gemini-2.5-flash for vision analysis. Returns text descriptions
instead of blocking, enabling Claude to "see" images via proxy.
Key changes:
- Add image-analyzer-transformer.cjs hook script
- Add ImageAnalysisConfig type and loader
- Add hook installer and profile injector
- Add prompt templates for analysis customization
- Add e2e test suite (excluded from normal CI runs)
- Configure test:e2e script for manual testing
Environment variables:
- CCS_IMAGE_ANALYSIS_ENABLED: Enable/disable (default: 1)
- CCS_IMAGE_ANALYSIS_MODEL: Vision model (default: gemini-2.5-flash)
- CCS_IMAGE_ANALYSIS_TIMEOUT: Timeout in seconds (default: 60)
- CCS_CLIPROXY_API_KEY: API key for CLIProxy auth
- CCS_CLIPROXY_PORT: CLIProxy port (default: 8317)
Closes#426
Previously, PUT /api/settings/:profile created a backup on every request
regardless of whether the content actually changed. This led to hundreds
of identical backup files accumulating in ~/.ccs/backups/.
Changes:
- Compare existing content with new content before creating backup
- Reuse computed newContent for atomic write (DRY)
- Make hook injection idempotent by checking content before write
Fixes#433
- Read ANTHROPIC_MODEL from profile settings instead of hardcoding
- Display model name in full uppercase (GLM-4.7, not Glm-4.7)
- Add null/undefined guard to getModelDisplayName
- Remove hardcoded GLM-4.6/GLM-4.6 (Thinking) display names
Closes#431
Match WebSearch hook pattern:
- ENABLED by default for settings/cliproxy profiles
- DISABLED for native Claude accounts (account/default)
- User can override via config: hooks.block_image_read.enabled: false
This ensures CCS CLI users get context protection out-of-the-box
while native Claude subscription users are unaffected.
Add PreToolUse hook that intercepts Read tool calls on image files
(.png, .jpg, .webp, etc.) and blocks them with helpful message.
This prevents context exhaustion when image generation skills
produce multiple files and the agent tries to read them (each
image can consume 100K+ tokens).
Configuration:
- Enable via config.yaml: hooks.block_image_read.enabled: true
- Or env var: CCS_BLOCK_IMAGE_READ=1
Hook integration:
- lib/hooks/block-image-read.cjs - the hook script
- src/utils/hooks/image-read-block-hook-env.ts - config loader
- Integrated into all spawn locations (ccs.ts, shell-executor,
cliproxy-executor)
Closes#426
Add 17 tests for isCcsWebSearchHook and deduplicateCcsHooks functions
covering Unix/Windows paths, edge cases, and duplicate cleanup.
Also add invariant comment in hook-config.ts.
Addresses code review feedback from PR #424.
- Create hook-utils.ts with isCcsWebSearchHook() and deduplicateCcsHooks()
- Update profile-hook-injector.ts and hook-config.ts to use shared module
- Combine double writeFileSync into single write in ensureHookConfig()
Addresses code review feedback on PR #420
- Add typeof check before .replace() to prevent crash on non-string commands
- Add isCcsWebSearchHook() and deduplicateCcsHooks() to hook-config.ts
- Fix ensureHookConfig() detection to verify path, not just matcher
- Prevent overwriting user's custom WebSearch hooks
On Windows, symlinks often fail so copyFallback() copies directories instead.
Previously, subsequent `ccs sync` runs would see the copied folder, fail the
isOurSymlink() check, and create a new .backup-{date} folder every time.
This fix adds Windows copy detection in installItem() before calling backupItem().
When a valid CCS copy is detected, it refreshes the content instead of backing up.
Fixes#409
Remove global hook registration from installWebSearchHook() - now only
copies hook file to ~/.ccs/hooks/. Hook registration handled by
ensureProfileHooks() which writes to per-profile settings.
Changes:
- hook-installer.ts: Remove ensureHookConfig() call
- ccs.ts: Remove redundant installWebSearchHook() call
- hook-config.ts: Add getClaudeSettingsPath() for test isolation
- postuninstall.js: add file logging for debugging on error
- profile-hook-injector.ts: use 'wx' flag for atomic marker creation
- profile-hook-injector.ts: include parse error message in debug log
- install-command.ts: use actual counts for consistent semantics
- Windows tests: align Section 7 with per-profile hook architecture
- ClaudeSymlinkManager.uninstall() now returns count for accurate reporting
- install-command.ts uses return value instead of always incrementing
- claude-dir-installer.ts uses getCcsHome() directly instead of string manipulation
- Add TODO comments to tests explaining --install is a no-op
Fix test isolation in symlink managers and checkers to prevent tests
from touching user's real ~/.ccs/ and ~/.claude/ directories.
Files fixed:
- src/api/services/profile-writer.ts - use getCcsDir()
- src/management/checks/symlink-check.ts - use functions instead of
module-level constants
- src/utils/claude-dir-installer.ts - use getCcsDir() for CCS paths
- src/utils/claude-symlink-manager.ts - use getCcsHome()/getCcsDir()
Replace os.homedir() with getCcsDir() in 11 source files to ensure
tests don't touch the user's real ~/.ccs/ directory. The getCcsDir()
function from config-manager.ts respects CCS_HOME env var for
test isolation.
Files fixed:
- src/auth/profile-detector.ts
- src/auth/profile-registry.ts
- src/delegation/headless-executor.ts
- src/delegation/session-manager.ts
- src/glmt/glmt-transformer.ts
- src/management/checks/config-check.ts
- src/management/checks/profile-check.ts
- src/management/checks/system-check.ts
- src/management/instance-manager.ts
- src/management/shared-manager.ts
- src/utils/update-checker.ts
Ensure all hook-related modules use getCcsDir() from environment.ts
for consistent test isolation. Prevents tests from touching the user's
real ~/.ccs/ directory during test runs.
Changes:
- hook-config.ts: Use getCcsDir() for hookConfigPath
- hook-installer.ts: Use getCcsDir() for HOOK_SOURCE_PATH
- profile-hook-injector.ts: Use getCcsDir() for hook sources
- CLAUDE.md: Add test isolation rules
- remove removeHookConfig() call from uninstallWebSearchHook()
- add removeMigrationMarker() cleanup
- update postuninstall.js to only clean CCS files
- global ~/.claude/settings.json is never touched
- ensureProfileHooks() injects hooks into profile settings
- migrateGlobalHook() one-time migration from global settings
- removeMigrationMarker() cleanup for uninstall
- export from barrel files for module access
* fix(doctor): use dynamic profile discovery for delegation check
Replace hardcoded ['glm', 'kimi'] list with DelegationValidator.getReadyProfiles()
to detect all configured *.settings.json profiles including mm, or1, g7, etc.
* fix(ci): exclude bot comments from triggering AI review
Bot progress comments were triggering new workflow runs,
which cancelled in-progress reviews due to concurrency group.
Added check for github.event.comment.user.type != 'Bot'.
* chore(release): 7.18.0-dev.1 [skip ci]
* fix(delegation): only check profiles defined in config.yaml
Previously getReadyProfiles() scanned all *.settings.json files,
including orphan files (ghcp, kiro) not in config.yaml.
Now reads from config.yaml:
- profiles section (excluding 'default')
- cliproxy.providers section
Fixes doctor showing 11 profiles instead of configured 9.
* chore(release): 7.18.0-dev.2 [skip ci]
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
- Clear stdinKeepalive interval on SIGINT/SIGTERM signal handlers
- Add cancelProjectSelection() to timeout/exit/error handlers
- Close server on error path in testLocalhostBinding to prevent fd leak
- Add TTL-based cleanup for stale auth sessions (10 min expiry)
- Use DEVICE_CODE_TIMEOUT_MS constant instead of hardcoded value (DRY)
- Extract common validation logic into validateProviderKey() function
- Convert validateGlmKey() and validateMiniMaxKey() to thin wrappers
- Remove 15+ unnecessary inline comments explaining obvious code
- Remove verbose JSDoc that duplicates function signatures
- Reduce file from 236 to 148 lines (37% reduction)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Restore src/commands/migrate-command.ts to dev branch state (was truncated)
- Remove src/config/rename-minimax-profile.ts (circular export causing build break)
- Fix mm.settings.json path typo in api-key-validator.ts
Build now passes: bun run typecheck && bun run lint:fix
* fix(cliproxy): pass variant port to executor for isolation
Variants configured with dedicated ports (8318-8417) were not using
their assigned port. The executor always defaulted to 8317.
Changes:
- Add port field to ProfileDetectionResult interface
- Pass variant.port from profile-detector to ccs.ts
- Forward port to execClaudeWithCLIProxy options
- Update executor priority: CLI flags > variant port > config.yaml > default
Closes#228
* fix(cliproxy): propagate port in unified config and UI preset handlers
Edge case fixes identified in codebase review:
- Unified config variant detection: add settingsPath and port fields
- Provider editor: use variant port in handleApplyPreset/handleCustomPresetApply
- preset-utils: add optional port parameter to applyDefaultPreset()
* chore(release): 7.11.1-dev.1 [skip ci]
* fix(cliproxy): use correct default port (8317) for remote HTTP connections
Root cause: Inconsistent default port logic across code paths.
- Test Connection used 8317 (correct)
- Actual API calls used 80 (wrong)
Changes:
- Add centralized getRemoteDefaultPort() helper in config-generator.ts
- Fix proxy-target-resolver.ts to use shared helper
- Fix rewriteLocalhostUrls() and getRemoteEnvVars() in config-generator.ts
- Update remote-proxy-client.ts to use shared helper (DRY)
Fixes all 3 reported issues:
1. Port empty → now correctly uses :8317 instead of :80
2. BASEURL construction → now includes correct port
3. CLIProxy Plus auth → now fetches from remote on correct port
* chore(release): 7.11.1-dev.2 [skip ci]
* feat(delegation): add Claude Code CLI flag passthrough
Add explicit passthrough support for key Claude Code CLI flags:
- --max-turns: Limit agentic turns (prevents infinite loops)
- --fallback-model: Auto-fallback when model overloaded
- --agents: Dynamic subagent JSON injection
- --betas: Enable experimental features
Maintain extraArgs catch-all for future Claude Code flags.
Update help command with new "Delegation Flags" section.
Closes#89
* test(delegation): add comprehensive CLI flag passthrough tests
Add 45 test cases covering all edge cases for CLI flag passthrough:
- DelegationHandler: timeout/max-turns/fallback-model/agents/betas validation
- HeadlessExecutor: duplicate flag filtering, undefined vs truthy checks
* chore(release): 7.11.1-dev.3 [skip ci]
* fix(ui): enable cancel button during OAuth authentication
Resolves#234 - Cancel button was disabled during authentication flow,
preventing users from canceling the OAuth process.
Changes:
- Add auth-session-manager.ts for tracking active OAuth sessions
- Add POST /cliproxy/auth/:provider/cancel endpoint to abort sessions
- Kill spawned CLIProxy auth process when cancel is triggered
- Enable Cancel button in AddAccountDialog during authentication
- Add cancel support to QuickSetupWizard auth step
- Update useCancelAuth hook to call backend cancel endpoint
* chore(release): 7.11.1-dev.4 [skip ci]
* fix(prompt): add stdin.pause() to prevent process hang after password input
Fixes#236. The password() method called resume() on stdin but never
paused it in cleanup, keeping the event loop alive indefinitely.
* chore(release): 7.11.1-dev.5 [skip ci]
* feat(cliproxy): add --allow-self-signed flag for HTTPS connections (#227)
Previously, allowSelfSigned was hardcoded to true for all HTTPS protocol
connections, forcing use of the native https module which has issues with
Cloudflare-proxied connections causing timeouts.
This change:
- Adds --allow-self-signed CLI flag (default: false)
- Adds CCS_ALLOW_SELF_SIGNED environment variable
- Uses standard fetch API by default for HTTPS (works with valid certs)
- Only uses native https module when --allow-self-signed is specified
Usage:
- For production HTTPS proxies with valid certs: no flag needed
- For dev proxies with self-signed certs: use --allow-self-signed
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* chore(release): 7.11.1-dev.6 [skip ci]
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Shun Kakinoki <39187513+shunkakinoki@users.noreply.github.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
- Wrap JSON.parse in try-catch with clear error message for malformed JSON
- Add iflow, kiro, ghcp providers to CLIProxyVariantConfig type
- Make settings optional in CLIProxyVariantConfig (was required with empty string fallback)
- Remove unsafe type cast in loadConfigSafe()
Addresses additional edge cases from PR #215 code review.
Fixes#214 - Fresh install fails with 'Config not found' when running
ccs config before RecoveryManager runs.
Changes:
- Move RecoveryManager.recoverAll() before all early-exit commands in ccs.ts
- Fix loadConfigSafe() to return empty config instead of throwing in legacy mode
- Add getActiveConfigPath() for mode-aware config path resolution
- Rename cliproxy getConfigPath() to getCliproxyConfigPath() to avoid confusion
- Update help-command.ts to show correct config path based on mode
This ensures all commands benefit from auto-recovery on fresh installs,
and gracefully handles missing config files in all code paths.
The try/catch around loadConfig() in getSettingsPath() was ineffective
because process.exit() cannot be caught by try/catch - it terminates
the process immediately.
Replace with inline safe logic that:
1. Checks if config.json exists with fs.existsSync()
2. Reads and parses JSON manually
3. Uses isConfig() type guard for validation
4. Properly catches JSON parse errors
This ensures unified mode users don't crash when falling back to check
legacy config.json for profiles not found in config.yaml.
Add loadConfigSafe() function that handles both unified (config.yaml) and
legacy (config.json) config formats. Uses throwable errors instead of
process.exit() so try/catch blocks work properly in web server routes.
Updated files to use loadConfigSafe():
- overview-routes.ts: Dashboard overview API
- route-helpers.ts: readConfigSafe() helper
- profile-reader.ts: API profile reading
- profile-writer.ts: API profile writing
- variant-config-adapter.ts: CLIProxy variant config
Fixes#206 (Problem 2: config.json not found when user has config.yaml)
- Add SIGKILL escalation after 3s SIGTERM timeout in session-tracker
- Auto-stop running proxy before binary update in binary-manager
- Add waitForPortFree utility to port-utils
- Detect version mismatch on startup and auto-restart outdated proxy
- Store and expose version in session lock for detection
Fixes issue where old proxy version continues running after update,
causing UI to show different version than actually running.
Linux kernel truncates process names to 15 chars, causing
'cli-proxy-api-plus' to appear as 'cli-proxy-api-p' in lsof output.
Switch from exact whitelist to prefix matching for robust detection.
Two bugs caused the health check to incorrectly report "Proxy not running":
1. lsof command used mutually exclusive flags (-t and -F)
- Removed -t flag since -F already provides structured output
2. isCLIProxyProcess whitelist was missing common binary names
- Added: cliproxyapi, cli-proxy-api-plus (and .exe variants)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
kaitranntt