603 Commits

Author SHA1 Message Date
YeonGyu-Kim
0bffdc441e feat(hooks): add sisyphus-gpt-hephaestus-reminder hook
Shows error toast when Sisyphus runs with a GPT model, nudging user to
use Hephaestus instead.

🤖 Generated with [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode)
2026-02-18 01:08:40 +09:00
YeonGyu-Kim
d0bd24bede fix(cli-run): rely on continuation markers for completion
Use hook-written continuation marker state to gate run completion checks and remove the noisy event-stream shutdown timeout log in run mode.
2026-02-17 17:50:47 +09:00
YeonGyu-Kim
f4eef9f534
Merge pull request #1907 from BowTiedSwan/fix/json-retry-loop
feat(hooks): add json-error-recovery hook to prevent infinite retry loops
2026-02-17 15:59:44 +09:00
YeonGyu-Kim
5f939f900a fix(hooks): harden json-error-recovery matching and scope 2026-02-17 15:46:21 +09:00
YeonGyu-Kim
462e2ec2b0 refactor: remove 3 orphaned files and prefix unused parameter 2026-02-17 14:09:12 +09:00
YeonGyu-Kim
aac79f03b5 docs: regenerate all AGENTS.md files from comprehensive codebase exploration
- Fired 33 parallel explore agents across all major directories
- Analyzed 1164 TS files, 133k LOC, 41 hooks, 26 tools, 11 agents, 18 features
- Regenerated 13 AGENTS.md files with 905 total lines
- Root: plugin architecture, initialization flow, 7 OpenCode hook handlers
- src/: entry point orchestration, hook composition pipeline
- agents/: 11 agent inventory with tool restrictions and factory patterns
- hooks/: 41 hooks organized by 5 tiers, key complex hooks documented
- tools/: 26 tools across 14 directories, delegation categories
- features/: 18 modules mapped by complexity (HIGH/MEDIUM/LOW)
- shared/: 101 utilities in 13 categories, model resolution pipeline
- config/: 22 schema files, Zod v4 validation system
- cli/: 5 commands, doctor checks, model fallback system
- mcp/: 3-tier MCP system architecture
- plugin-handlers/: 6-phase config loading pipeline
- claude-code-hooks/: CC settings.json compatibility layer
- claude-tasks/: task schema + file-based persistence

🤖 Generated with assistance of oh-my-opencode
2026-02-17 11:17:01 +09:00
YeonGyu-Kim
d786691260 fix: read anthropic 1m flag from live model cache state 2026-02-17 10:51:01 +09:00
YeonGyu-Kim
363016681b test: cover model-cache and env fallback context limits 2026-02-17 10:51:01 +09:00
YeonGyu-Kim
b444899153 fix: use model cache context flag for runtime context limits 2026-02-17 10:51:01 +09:00
bowtiedswan
86f2a93fc9
feat(hooks): add json-error-recovery hook to prevent infinite retry loops 2026-02-16 21:35:58 +02:00
YeonGyu-Kim
2b5887aca3 fix: prevent overlapping poll cycles in managers
Guarding polling re-entry avoids stacked async polls under slow responses, and unref on pending-call cleanup timer reduces idle wakeups.
2026-02-17 03:06:40 +09:00
YeonGyu-Kim
c44509b397 fix: skip startup toasts in CLI run mode for auto-update-checker
Add OPENCODE_CLI_RUN_MODE environment variable check to skip all startup
toasts and version checks when running in CLI mode. This prevents
notification spam during automated CLI run sessions.

Includes comprehensive test coverage for CLI run mode behavior.

🤖 Generated with OhMyOpenCode assistance
2026-02-17 02:34:39 +09:00
YeonGyu-Kim
b7497d0f9f
Merge branch 'dev' into fix/1700-vertex-anthropic 2026-02-17 01:54:11 +09:00
YeonGyu-Kim
d672eb1c12 fix: recognize google-vertex-anthropic as Claude provider (#1700) 2026-02-17 01:28:27 +09:00
YeonGyu-Kim
af7b1ee620 refactor(hashline): override native edit tool instead of separate tool + disabler hook
Replace 3-component hashline system (separate hashline_edit tool + edit
disabler hook + OpenAI-exempted read enhancer) with 2-component system
that directly overrides the native edit tool key, matching the
delegate_task pattern.

- Register hashline tool as 'edit' key to override native edit
- Delete hashline-edit-disabler hook (no longer needed)
- Delete hashline-provider-state module (no remaining consumers)
- Remove OpenAI exemption from read enhancer (explicit opt-in means all providers)
- Remove setProvider wiring from chat-params
2026-02-17 00:03:10 +09:00
YeonGyu-Kim
51dde4d43f feat(hashline): port hashline edit tool from oh-my-pi
This PR ports the hashline edit tool from oh-my-pi to oh-my-opencode as an experimental feature.

## Features
- New experimental.hashline_edit config flag
- hashline_edit tool with 4 operations: set_line, replace_lines, insert_after, replace
- Hash-based line anchors for safe concurrent editing
- Edit tool disabler for non-OpenAI providers
- Read output enhancer with LINE:HASH prefixes
- Provider state tracking module

## Technical Details
- xxHash32-based 2-char hex hashes
- Bottom-up edit application to prevent index shifting
- OpenAI provider exemption (uses native apply_patch)
- 90 tests covering all operations and edge cases
- All files under 200 LOC limit

## Files Added/Modified
- src/tools/hashline-edit/ (7 files, ~400 LOC)
- src/hooks/hashline-edit-disabler/ (4 files, ~200 LOC)
- src/hooks/hashline-read-enhancer/ (3 files, ~400 LOC)
- src/features/hashline-provider-state.ts (13 LOC)
- src/config/schema/experimental.ts (hashline_edit flag)
- src/config/schema/hooks.ts (2 new hook names)
- src/plugin/tool-registry.ts (conditional registration)
- src/plugin/chat-params.ts (provider state tracking)
- src/tools/index.ts (export)
- src/hooks/index.ts (exports)
2026-02-17 00:03:10 +09:00
YeonGyu-Kim
158ca3f22b refactor: remove unused params/imports/types from lsp-tools, task-tools, delegate-task, skill-loader, context-window-monitor, plugin-config 2026-02-16 22:12:21 +09:00
YeonGyu-Kim
9dbb9552b8 refactor: remove unused imports from auto-update-checker, claude-code-hooks, mcp 2026-02-16 22:11:38 +09:00
YeonGyu-Kim
bfabad7681 refactor: remove unused imports from interactive-bash-session, session-recovery, start-work 2026-02-16 22:11:35 +09:00
YeonGyu-Kim
169c07ebf8 refactor: remove unused imports from injector, tool-result-storage-sdk, session-notification-utils, model-resolver 2026-02-16 22:11:05 +09:00
YeonGyu-Kim
8dd3d07efd refactor: remove unused hasIgnoredParts variables from context-window-limit-recovery 2026-02-16 22:10:44 +09:00
YeonGyu-Kim
560d13dc70 Normalize agent name comparisons to handle display name keys
Hooks and tools now use getAgentConfigKey() to resolve agent names (which may
be display names like 'Atlas (Plan Executor)') to lowercase config keys
before comparison.

- session-utils: orchestrator check uses getAgentConfigKey
- atlas event-handler: boulder agent matching uses config keys
- category-skill-reminder: target agent check uses config keys
- todo-continuation-enforcer: skipAgents comparison normalized
- subagent-resolver: resolves 'metis' -> 'Metis (Plan Consultant)' for lookup
2026-02-16 20:43:09 +09:00
YeonGyu-Kim
66e66e5d73 test: add tests for SDK recovery modules (empty-content-recovery, recover-empty-content-message) 2026-02-16 18:20:32 +09:00
YeonGyu-Kim
1a6810535c refactor: create normalizeSDKResponse helper and replace scattered patterns across 37 files 2026-02-16 18:20:19 +09:00
YeonGyu-Kim
eb6067b6a6 fix: rename prompt_async to promptAsync for SDK compatibility 2026-02-16 16:52:06 +09:00
YeonGyu-Kim
c1681ef9ec fix: normalize SDK response shape in readMessagesFromSDK
Use response.data ?? response to handle both object and array-shaped
SDK responses, consistent with all other SDK readers.
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
9889ac0dd9 fix: handle array-shaped SDK responses in getSdkMessages & dedup getMessageDir
- getSdkMessages now handles both response.data and direct array
  responses from SDK
- Consolidated getMessageDir: storage.ts now re-exports from shared
  opencode-message-dir.ts (with path traversal guards)
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
5a6a9e9800 fix: defensive SDK response handling & parts-reader normalization
- Replace all response.data ?? [] with (response.data ?? response)
  pattern across 14 files to handle SDK array-shaped responses
- Normalize SDK parts in parts-reader.ts by injecting sessionID/
  messageID before validation (P1: SDK parts lack these fields)
- Treat unknown part types as having content in
  recover-empty-content-message-sdk.ts to prevent false placeholder
  injection on image/file parts
- Replace local isRecord with shared import in parts-reader.ts
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
8edf6ed96f fix: address 5 SDK compatibility issues from Cubic round 8
- P1: Use compacted timestamp check instead of nonexistent truncated
  field in target-token-truncation.ts
- P1: Use defensive (response.data ?? response) pattern in
  hook-message-injector/injector.ts to match codebase convention
- P2: Filter by tool type in countTruncatedResultsFromSDK to avoid
  counting non-tool compacted parts
- P2: Treat thinking/meta-only messages as empty in both
  empty-content-recovery-sdk.ts and message-builder.ts to align
  SDK path with file-based logic
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
cfb8164d9a docs: regenerate all 13 AGENTS.md files from deep codebase exploration 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
c2012c6027 fix: address 8-domain Oracle review findings (C1, C2, M1-M4)
- C1: thinking-prepend unique part IDs per message (global PK collision)
- C2: recover-thinking-disabled-violation try/catch guard on SDK call
- M1: remove non-schema truncated/originalSize fields from SDK interfaces
- M2: messageHasContentFromSDK treats thinking-only messages as non-empty
- M3: syncAllTasksToTodos persists finalTodos + no-id rename dedup guard
- M4: AbortSignal.timeout(30s) on HTTP fetch calls in opencode-http-api

All 2739 tests pass, typecheck clean.
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
106cd5c8b1 fix: re-read fresh messages before empty scan & dedup isRecord import
- Re-read messages from SDK after injectTextPartAsync to prevent stale
  snapshot from causing duplicate placeholder injection (P2)
- Replace local isRecord with shared import from record-type-guard (P3)
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
c799584e61 fix: address Cubic round-6 P2/P3 issues
- P2: treat unknown part types as non-content in message-builder messageHasContentFromSDK
- P3: reuse shared isRecord from record-type-guard.ts in opencode-http-api
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
3fe9c1f6e4 fix: address Cubic round-5 P1/P2 issues
- P1: add path traversal guard to getMessageDir (reject .., /, \)
- P2: treat unknown part types as non-content in messageHasContentFromSDK
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
885c8586d2 fix: revert messageHasContentFromSDK unknown type handling
Unknown part types should be treated as content (return true)
to match parity with the existing message-builder implementation.
Using continue would incorrectly mark messages with unknown part
types as empty, triggering false recovery.
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
8d82025b70 fix: address Cubic round-4 P2 issues
- isTodo: allow optional id to match Todo interface, preventing
  todos without ids from being silently dropped
- messageHasContentFromSDK: treat unknown part types as empty
  (continue) instead of content (return true) for parity with
  existing storage logic
- readMessagesFromSDK in recover-empty-content-message-sdk: wrap
  SDK call in try/catch to prevent recovery from throwing
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
557340af68 fix: restore readMessagesFromSDK and its test
The previous commit incorrectly removed this function and its test
as dead code. While the local implementations in other files have
different return types (MessageData[], MessagePart[]) and cannot be
replaced by this shared version, the function is a valid tested
utility. Deleting tests is an anti-pattern in this project.
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
d7b38d7c34 fix: address Cubic round-3 P2/P3 issues
- Encode path segments with encodeURIComponent in HTTP API URLs
  to prevent broken requests when IDs contain special characters
- Remove unused readMessagesFromSDK from messages-reader.ts
  (production callers use local implementations; dead code)
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
880b53c511 fix: address Cubic round-2 P2 issues
- target-token-truncation: eliminate redundant SDK messages fetch by
  extracting tool results from already-fetched toolPartsByKey map
- recover-thinking-block-order: wrap SDK message fetches in try/catch
  so recovery continues gracefully on API errors
- thinking-strip: guard against missing part.id before calling
  deletePart to prevent invalid HTTP requests
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
3bbe0cbb1d feat: implement SDK/HTTP pruning for deduplication and tool-output truncation on SQLite
- executeDeduplication: now async, reads messages from SDK on SQLite via
  client.session.messages() instead of JSON file reads
- truncateToolOutputsByCallId: now async, uses truncateToolResultAsync()
  HTTP PATCH on SQLite instead of file-based truncateToolResult()
- deduplication-recovery: passes client through to both functions
- recovery-hook: passes ctx.client to attemptDeduplicationRecovery

Removes the last intentional feature gap on SQLite backend — dynamic
context pruning (dedup + tool-output truncation) now works on both
JSON and SQLite storage backends.
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
52161ef69f fix: add SDK readParts fallback for recoverToolResultMissing on SQLite
On SQLite backend, readParts() returns [] since JSON files don't exist.
Add isSqliteBackend() branch that reads parts from SDK via
client.session.messages() when failedAssistantMsg.parts is empty.
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
62e4e57455 feat: wire context-window-recovery callers to async SDK/HTTP variants on SQLite
- empty-content-recovery: isSqliteBackend() branch delegating to extracted
  empty-content-recovery-sdk.ts with SDK message scanning
- message-builder: sanitizeEmptyMessagesBeforeSummarize now async with SDK path
  using replaceEmptyTextPartsAsync/injectTextPartAsync
- target-token-truncation: truncateUntilTargetTokens now async with SDK path
  using findToolResultsBySizeFromSDK/truncateToolResultAsync
- aggressive-truncation-strategy: passes client to truncateUntilTargetTokens
- summarize-retry-strategy: await sanitizeEmptyMessagesBeforeSummarize
- client.ts: derive Client from PluginInput['client'] instead of manual defs
- executor.test.ts: .mockReturnValue() → .mockResolvedValue() for async fns
- storage.test.ts: add await for async truncateUntilTargetTokens
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
dff3a551d8 feat: wire session-recovery callers to async SDK/HTTP variants on SQLite
- recover-thinking-disabled-violation: isSqliteBackend() branch using
  stripThinkingPartsAsync() with SDK message enumeration
- recover-thinking-block-order: isSqliteBackend() branch using
  prependThinkingPartAsync() with SDK orphan thinking detection
- recover-empty-content-message: isSqliteBackend() branch delegating to
  extracted recover-empty-content-message-sdk.ts (200 LOC limit)
- storage.ts barrel: add async variant exports for all SDK functions
2026-02-16 16:13:40 +09:00
YeonGyu-Kim
0a085adcd6 fix(test): rewrite SDK reader tests to use mock client objects instead of mock.module 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
2bf8b15f24 feat: migrate hook callers to SDK message finders on SQLite backend 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
af8de2eaa2 feat: add SDK read paths for session-recovery parts/messages readers 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
1197f919af feat: add SDK/HTTP paths for tool-result-storage truncation 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
808de5836d feat: implement SQLite backend for replaceEmptyTextParts via HTTP PATCH 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
f69820e76e feat: implement SQLite backend for prependThinkingPart via HTTP PATCH 2026-02-16 16:13:40 +09:00
YeonGyu-Kim
c771eb5acd feat: implement SQLite backend for injectTextPart via HTTP PATCH 2026-02-16 16:13:40 +09:00