Files

T

YeonGyu-Kim 838b5ae216 docs(agents): refresh hierarchical AGENTS.md with team-mode coverage

Update root + 43 directory-level AGENTS.md files to reflect current state:
- Root AGENTS.md rewritten with accurate counts (1967 TS files, 1304 source +
  663 test, 278k LOC, 120 barrel index.ts), 7-step init flow, 5-tier hook
  composition, and full Team Mode section (12 team_* tools, eligibility,
  storage layout, config gate)
- src/AGENTS.md adds team-mode init step, current per-subdir file/LOC table
- src/tools/AGENTS.md documents conditional gates (team-mode +12, task
  system +4, hashline +1, interactive_bash +1, look_at +1) with always-on
  baseline of 20
- src/hooks/AGENTS.md splits into 5 tiers + 4 conditional team-mode hooks
- src/features/team-mode/AGENTS.md surfaces 12 tools, eligible agents,
  spawn-race-safe invariants, and integration points
- src/features/builtin-skills/AGENTS.md tracks 10 skills incl. team-mode
- src/agents/AGENTS.md, src/plugin/AGENTS.md, src/config/AGENTS.md updated
  for team-mode awareness, accurate counts, and current schema field list
- All other AGENTS.md files refreshed to 2026-05-08 generation date

2026-05-08 12:08:42 +09:00

2.9 KiB

Raw Blame History

src/hooks/keyword-detector/ — Mode Keyword Injection

Generated: 2026-05-08

OVERVIEW

Transform Tier hook on messages.transform. Scans first user message for mode keywords (ultrawork, search, analyze, team) and injects mode-specific system prompts.

KEYWORDS

Keyword	Pattern	Effect
`ultrawork` / `ulw`	`/\b(ultrawork	ulw)\b/i`
Search mode	`SEARCH_PATTERN` (from `search/`)	Web/doc search focus prompt injection
Analyze mode	`ANALYZE_PATTERN` (from `analyze/`)	Deep analysis mode prompt injection
Team mode	`TEAM_PATTERN` (from `team/`)	Forces orchestration via `team_*` tools when user invokes `team mode` / `팀 모드` / `팀으로`; instructs user to enable `team_mode.enabled` if tools are absent

STRUCTURE

keyword-detector/
├── index.ts           # Barrel export
├── hook.ts            # createKeywordDetectorHook() — chat.message handler
├── detector.ts        # detectKeywordsWithType() + extractPromptText()
├── constants.ts       # KEYWORD_DETECTORS array, re-exports from submodules
├── types.ts           # KeywordDetector, DetectedKeyword types
├── ultrawork/
│   ├── index.ts
│   ├── message.ts     # getUltraworkMessage() — dynamic prompt by agent/model
│   └── isPlannerAgent.ts
├── search/
│   ├── index.ts
│   ├── pattern.ts     # SEARCH_PATTERN regex
│   └── message.ts     # SEARCH_MESSAGE
├── analyze/
│   ├── index.ts
│   └── default.ts     # ANALYZE_PATTERN + ANALYZE_MESSAGE
└── team/
    ├── index.ts
    └── default.ts     # TEAM_PATTERN + TEAM_MESSAGE

DETECTION LOGIC

chat.message (user input)
  → extractPromptText(parts)
  → isSystemDirective? → skip
  → removeSystemReminders(text)  # strip <SYSTEM_REMINDER> blocks
  → detectKeywordsWithType(cleanText, agentName, modelID, disabledKeywords)
  → isPlannerAgent(agentName)? → filter out ultrawork
  → for each detected keyword: inject mode message into output

CONFIG

{
  "keyword_detector": {
    // Skip injection for any keyword in this list. Allowed: "ultrawork", "search", "analyze", "team".
    "disabled_keywords": ["search", "analyze"]
  }
}

Default: empty/missing → all four detectors active. Schema lives at src/config/schema/keyword-detector.ts.

GUARDS

System directive skip: Messages tagged as system directives are not scanned (prevents infinite loops)
Planner agent filter: Prometheus/plan agents do not receive ultrawork injection
Session agent tracking: Uses getSessionAgent() to get actual agent (not just input hint)
Model-aware messages: getUltraworkMessage(agentName, modelID) adapts message to active model

2.9 KiB Raw Blame History