# Skill: Tool & Skill Request Workflow

**Purpose:** Formal process for agents to request new tools or skills, with approval gate before creation.
**Owner:** Alfred (COO) — responsible for reviewing, escalating, creating, and notifying.
**Approval Authority:** Clayton (CEO)
**Channel:** LeadsPanther Team Telegram group → Clayton's Tasks topic

---

## When to Use

An agent should invoke this workflow when:
- They need a capability (tool, API, skill) that doesn't currently exist
- They discover a gap during task execution that blocks or degrades their work
- They identify a pattern that should be formalized as a reusable skill

## Request Format

The requesting agent must provide a structured request with ALL of the following:

```
TOOL/SKILL REQUEST
==================
Requesting Agent: [name] ([department])
Request Type: [tool | skill | api_access]
Name: [proposed name for the tool/skill]
Priority: [critical | high | medium | low]
Description: [what it does, 2-3 sentences]
Justification: [why it's needed, what task is blocked/degraded]
Benefiting Agents: [which agents would use this]
Existing Alternatives: [what was tried, why it's insufficient]
Estimated Complexity: [simple | moderate | complex]
```

## Workflow Steps

### 1. Agent identifies need
Agent recognizes a missing capability during task execution. Agent formats the request using the template above.

### 2. Agent sends request to Alfred
Agent posts the structured request in the LeadsPanther Team Telegram group, tagging Alfred.

**If the agent cannot post directly:** Write the request to `~/.openclaw/workspace/ops/requests/YYYY-MM-DD-[name].md` and Alfred will pick it up on next heartbeat.

### 3. Alfred reviews
Alfred checks:
- Does an existing skill/tool already solve this? (Check `~/.openclaw/skills/` directory)
- Is the request well-scoped and justified?
- Is the complexity estimate reasonable?
- Which agents benefit and is the impact worth the effort?

### 4. Alfred escalates to Clayton
Alfred posts a summary in **Clayton's Tasks** topic in the LeadsPanther Team Telegram group:

```
🔧 TOOL/SKILL REQUEST — [Priority]
From: [Agent] ([Department])
Request: [Name] — [one-line description]
Recommendation: APPROVE / DEFER / DENY
Reason: [Alfred's assessment in 1-2 sentences]
Complexity: [simple/moderate/complex]
```

### 5. Clayton decides
Clayton responds in Telegram with one of:
- **✅ Approved** — Alfred proceeds to create
- **⏸️ Deferred** — Revisit later (Alfred logs and schedules reminder)
- **❌ Denied** — Alfred informs requesting agent with reason

### 6. Alfred creates (if approved)
Alfred builds the tool or skill:

**For skills:**
- Create directory: `~/.openclaw/skills/[skill-name]/`
- Write `SKILL.md` with full documentation, examples, and usage instructions
- Follow existing skill format conventions

**For tools/API access:**
- Provision credentials if needed (add to `.env`)
- Update relevant agent's `TOOLS.md`
- Create usage documentation

**For both:**
- Update `knowledge/areas/skill-tracker.md` with new entry
- Update relevant agent's `SERVICES.md` if applicable

### 7. Alfred notifies requesting agent
Alfred posts in the team Telegram group:

```
✅ NEW [TOOL/SKILL]: [name]
Location: ~/.openclaw/skills/[name]/SKILL.md
Created for: [requesting agent]
Usage: [brief 1-line instruction]
```

### 8. Agent resumes
The requesting agent picks up the new capability and continues their work. Agent should confirm functionality and report any issues back to Alfred.

## Request Log

All requests should be tracked in `~/.openclaw/workspace/ops/requests/` with the following naming convention:
`YYYY-MM-DD-[request-name].md`

Status is tracked in the file header:
```
Status: pending | approved | denied | deferred | completed
```

## Emergency Requests

For **critical** priority requests that block revenue-generating work:
1. Agent posts in Clayton's Tasks topic directly (skip Alfred review)
2. Clayton can approve directly
3. Alfred creates immediately upon approval
4. Standard documentation follows after creation