Skip to content
Back to skills

Explore

ASecurity

Progressive codebase exploration: index, then signatures, then targeted reads only.

  • 5 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 2, 2026
ai-agentsbash

Security analysis

A100/100

Scanned September 2, 2026

npx -y skills add deepaksinghcs14/deadeye-cc --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/deepaksinghcs14-explore/badge)](https://www.skillsdirectory.com/skills/deepaksinghcs14-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: Progressive codebase exploration: index, then signatures, then targeted reads only.
context: fork
agent: Explore
---

Explore the codebase in three layers, stopping at the first one that
answers the question. Do not skip ahead to reading full files "to be safe"
-- each layer is a decision point, not a formality.

**Layer 1 -- file index.** `Glob`/`find` only, zero file reads. Get the
shape: directory layout, naming conventions, where the relevant area
likely lives. Stop here if the task is now obvious (e.g. "add a field to
this one struct" and you can see exactly which file that is).

**Layer 2 -- entry points and signatures.** `Grep` for the symbols,
exports, and call sites that matter to the task. You're building a map of
what calls what, not reading implementations yet. Stop here if you now
know which functions/files the change touches and roughly how they fit
together.

**Layer 3 -- targeted reads.** `Read` with `offset`/`limit` only for the
files directly relevant to the upcoming change -- never a full-repo read
sweep. If a file is large, read the section around the relevant symbol,
not the whole thing.

After each layer, explicitly decide: enough context to proceed, or one
more layer? Don't default to Layer 3 out of caution -- most tasks resolve
at Layer 1 or 2.

Return a structured summary to the caller: what you found, which
files/symbols matter, and your recommendation -- not the raw contents you
read along the way. The raw reads stay in this forked context; only the
summary crosses back.

**Before returning, cache the summary** so the next session doesn't
re-derive it (best-effort -- if this fails in any way, skip it silently
and return your summary as normal; it is a cache, never a deliverable):

```bash
deadeye notes-append explore <<'EOF'
<the question you were asked>
- <a file or symbol that matters, and why>
- recommendation: <one line>
EOF
```

Ten lines maximum. If `deadeye` isn't on PATH, retry once with
`~/.deadeye/bin/deadeye`; if that also fails, move on.

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.