Skip to content
Back to skills

Explore

ASecurity

Time-boxed SENAR investigation before a task.

  • 18 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 6, 2026
ai-agentsgo

Works with

  • mcp

Security analysis

A100/100

Scanned October 6, 2026

npx -y skills add Kibertum/tausik-core --skill explore --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Explore?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Explore
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/kibertum-explore/badge)](https://www.skillsdirectory.com/skills/kibertum-explore)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
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

1. **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.

2. **Announce:**
   - What we're investigating
   - Time limit
   - What we hope to learn

3. **Investigate:**
   - **Locate code through RAG** — `mcp__codebase-rag__search_code` for concepts; a `Grep` also brings the index's top chunks automatically.
   - Skim what the search returns, then `Read` the specific files that look relevant.
   - Try small experiments (do NOT write production code).
   - Document findings as you go in conversation.

4. **Check time periodically** via `tausik_explore_current` MCP tool. If over time limit — wrap up immediately.

5. **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.

6. **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"`.

7. **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.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.