# Zephex

> >-

- **Type:** Skill
- **Install:** `agentstack add skill-zephexmcp-agent-skills-zephex`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [zephexMCP](https://agentstack.voostack.com/s/zephexmcp)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [zephexMCP](https://github.com/zephexMCP)
- **Source:** https://github.com/zephexMCP/agent-skills/tree/main/skills/zephex
- **Website:** https://zephex.dev

## Install

```sh
agentstack add skill-zephexmcp-agent-skills-zephex
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Zephex MCP

Ten tools at `https://zephex.dev/mcp` for **the user's project** — not training data.

They may only say *"use Zephex"* or *"use MCP."* You choose the tool and call it (`find_code`, `read_code`, …). Some editors show `zephex:find_code` — same tool.

## Workflow

| Situation | Chain |
|-----------|--------|
| New repo | `get_project_context` → optional `explain_architecture` |
| Feature, fix, refactor | `scope_task` → `read_code` on `focus_files` |
| Unknown location | `find_code` → `read_code` |
| Large rename | `find_code` + `exhaustive: true` → `files_summary` |
| Add or upgrade dependency | `check_package` → `audit_package` if needed |
| Stuck after 2+ tries | `keep_thinking` ↔ `find_code` / `read_code` |
| Production URL | `audit_headers` |
| Generic patterns (not their repo) | `Zephex_dev_info` search → get |

**`path`** on every repo tool: absolute folder (`/Users/jane/api`) or `github:owner/repo`. Monorepo: `path` = app root, not empty parent.

**`inline_files`:** only without disk access — `{ "rel/path": "" }`.

Max **3** calls of the same tool per turn without new evidence.

### Step 1 — Orient or scope

New repo or vague ask:

```
get_project_context({ "path": "/abs/project", "detail_level": "brief" })
```

Before reading many files for a task:

```
scope_task({
  "task": "Add rate limiting to POST /api/upload",
  "path": "/abs/project",
  "hint_symbols": ["handleUpload"]
})
```

Use `focus_files`, `callers_at_risk`, `suggested_approach`. Good `task`: one clear sentence. Bad: `"fix it"`.

### Step 2 — Read or search

Known symbol:

```
read_code({ "target": "handleUpload", "path": "/abs/project" })
```

Unknown location:

```
find_code({ "query": "validateToken", "path": "/abs/project", "scope": "definitions" })
```

Then `read_code` on the best hit. Skip `find_code` if you already know the symbol.

### Step 3 — Package, architecture, URL, reasoning

```
check_package({ "package": "express", "ecosystem": "npm" })
```

```
explain_architecture({ "path": "/abs/project", "focus": "auth" })
```

```
audit_headers({ "url": "https://example.com" })
```

```
keep_thinking({
  "thought": "Hypothesis: rate limit runs after handler",
  "thoughtNumber": 1,
  "totalThoughts": 5,
  "nextThoughtNeeded": true,
  "confidence": 0.6,
  "thoughtType": "hypothesis",
  "goalAnchor": "Add rate limiting to POST /api/upload",
  "lastActions": ["find_code(query=rateLimit)"]
})
```

Tool name is **`keep_thinking`**, not `thinking`.

```
Zephex_dev_info({ "operation": "search", "query": "Stripe webhook signature" })
```

Then `operation: "get"` with the slug from search.

## Tool reference

| Tool | When | Key args |
|------|------|----------|
| `get_project_context` | New session | `path`, `force?` |
| `scope_task` | Task before bulk reads | `task`, `path`, `hint_symbols?`, `max_files?` (1–15) |
| `find_code` | Don't know where | `query`, `path`, `scope?`, `exhaustive?` |
| `read_code` | Known symbol/files | `target` or `mode`+`files`, `path` |
| `check_package` | Before install | `package`, `ecosystem` |
| `audit_package` | Upgrade/CVE | `package`, `task`, `from_version?` |
| `explain_architecture` | How this repo works | `path`, `focus?` |
| `audit_headers` | User gave URL | `url` |
| `keep_thinking` | Stuck / risky change | `thought`, `thoughtNumber`, `goalAnchor`, `lastActions` |
| `Zephex_dev_info` | Playbooks | `operation`: search → get |

`read_code` `callers` / `blast_radius`: local `path` only — on `github:` use `find_code` usages.

## If they only said "use Zephex"

1. Guess the goal from their message.
2. Pick a **Workflow** row.
3. Call with `path` + specific args.
4. Answer from tool JSON only.

## Errors

| Error | Action |
|-------|--------|
| Unauthorized | `npx zephex setup` |
| 429 | Tell user; retry once |
| Missing `path` | Ask absolute project directory |

## Mistakes

- Install without `check_package`
- Many reads without `scope_task`
- `find_code` when `target` is known
- `thinking` instead of `keep_thinking`
- Repo answers from memory when Zephex is connected

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [zephexMCP](https://github.com/zephexMCP)
- **Source:** [zephexMCP/agent-skills](https://github.com/zephexMCP/agent-skills)
- **License:** MIT
- **Homepage:** https://zephex.dev

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-zephexmcp-agent-skills-zephex
- Seller: https://agentstack.voostack.com/s/zephexmcp
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.