| name | explore |
| description | Time-boxed SENAR investigation before a task. |
| context | fork |
| effort | medium |
/explore — Time-Bounded Exploration (SENAR Section 5.1)
Investigation without full task formality.
When to use: Unfamiliar domain, multiple possible approaches, need to understand before planning.
Key rule: If exploration yields implementation work, a task SHALL be created before changes are committed.
Model hint (phase=research): simple, read-only discovery (where-is-X,
symbol/pattern search, doc skim) runs fine on Haiku 4.5 — 10-20× cheaper.
Escalate to Sonnet 4.6 only for deep, multi-source investigation.
Rationale: docs/ru/research/model-routing-matrix.md.
Argument Dispatch
$ARGUMENTS = description of what to investigate
-
Start exploration via tausik_explore_start with title="What we're investigating" and time_limit=30 (integer, minutes, default 30). Use shorter for simple questions, longer for complex domains.
-
Announce:
- What we're investigating
- Time limit
- What we hope to learn
-
Investigate:
- Locate code via
mcp__codebase-rag__search_code first — it returns ranked chunks, not full files, and is the cheapest way to find symbols/patterns. Use Grep only for known file paths or when RAG is empty/stale; use Read only when you already have an exact path.
- Skim the chunks RAG returns, then
Read the specific files that look relevant.
- Try small experiments (do NOT write production code).
- Document findings as you go in conversation.
-
Check time periodically via tausik_explore_current MCP tool. If over time limit — wrap up immediately.
-
End exploration with findings via tausik_explore_end with summary="What was discovered" and optionally create_task=true to auto-create a task from findings.
-
Document dead ends via tausik_dead_end with approach="What was tried", reason="Why it failed". Check existing: tausik_memory_list with type="dead_end".
-
Suggest next: "Findings recorded. Create a task with /plan or continue exploring."
$ARGUMENTS = "end"
End current exploration via tausik_explore_end with summary="Findings summary".
$ARGUMENTS = "status" or empty
Check current exploration via tausik_explore_current (no params).
Rules
- Time-bounded. Respect the time limit. When it's up, stop and summarize.
- No production code. Explorations are for learning, not implementing. Small experiments only.
- If it yields work → create task. Use
--create-task or follow up with /plan.
- Document everything. Dead ends, findings, decisions — all captured.
- Cheap to start, cheap to end. Don't overthink the exploration scope.
Gotchas
- Only one active exploration at a time. End the current one before starting another.
- Explorations are not tasks — no plan steps, no QG-0, no AC. Just investigate and report.
--create-task requires --summary — you can't create a task without explaining what you found.