Scout Agent Skill

COG Scout Skill

Purpose

Lightweight URL/tool triage that sits between "ignore" and /url-dump. Evaluates whether a URL or tool is worth saving or skipping — checking existing vault coverage, assessing relevance to the user's profile and interests, and recommending a clear next action.

When to Invoke

• User wants to evaluate a URL or tool before committing to a full save
• User says "scout this", "evaluate this", "should I save this?", "is this relevant?"
• User shares one or more URLs and wants a quick relevance assessment
• User mentions a tool/service name and wants to know if it's worth investigating

Agent Mode Awareness

Check agent_mode in 00-inbox/MY-PROFILE.md frontmatter:

• If agent_mode: team — delegate vault scanning and web fetching to parallel sub-agents (one for vault search, one for content fetch/analysis). Combine results for recommendation.
• If agent_mode: solo (default) — handle all scanning and analysis directly in the conversation. No delegation.

Pre-Flight Check

Before executing, check for user profile:

• Look for 00-inbox/MY-PROFILE.md and 00-inbox/MY-INTERESTS.md in the vault
• If NOT found:

   Welcome to COG! Scout works best with a profile for relevance matching.

   Would you like to run /onboarding first, or should I evaluate with general criteria?

• If found:
• Read MY-PROFILE.md for active projects and role
• Read MY-INTERESTS.md for topic areas
• Read 00-inbox/MY-INTEGRATIONS.md for active integrations (check if web-fetch and web-search are available)

Boundary with /url-dump

Scout evaluates ("should I save this?"). URL-dump saves ("save this now").

• Scout checks existing coverage, assesses relevance, and recommends an action
• If the recommendation is Save, scout hands off to /url-dump with pre-filled category
• Users who already know they want to save should use /url-dump directly

Process Flow

1. Accept Input

Accept one or more of:

• URL(s): Direct links to evaluate
• Tool/service name(s): Will search for the tool first
• Mixed: Combination of URLs and names

Prompt (if no input provided):

What URL(s) or tool(s) would you like me to evaluate?
(You can paste URLs, tool names, or a mix)

Batch mode: Multiple URLs/names in one invocation are processed together with a summary table at the end.

2. Vault Coverage Check

For each URL or tool name, search the entire vault for existing coverage.

Search strategy:

• Extract domain from URL (e.g., github.com/owner/repo → search for repo name)
• Search for tool/service name across the whole vault (grep for domain, repo name, tool name)
• Match against URL strings in frontmatter (url: fields) and inline links

If found:

🔍 Existing coverage found for [name]:
- [file path] — saved [date], category: [category]
- [file path] — mentioned in [context]

Want me to check if an update is needed, or skip this one?

3. Content Fetch & Analysis

If URL provided and web-fetch is active:

• Fetch the URL content using WebFetch
• Extract: title, description, content type, author, date

If tool name provided (no URL):

• Use WebSearch to find the tool's primary page
• Fetch and analyze the top result

Content type detection:

• Tool/Service: Software, SaaS, API, library, framework
• Article/Blog: Long-form content, tutorial, opinion piece
• Repository: GitHub/GitLab repo (extract stars, last commit, language)
• Research: Paper, study, academic content
• News: Industry news, announcement
• Reference: Documentation, spec, standard

4. Relevance Assessment

Score relevance against user context:

Profile Match (from MY-PROFILE.md):

• Does it relate to an active project? Which one?
• Does it align with the user's role?
• Does it fit the user's tech stack?

Interest Match (from MY-INTERESTS.md):

• Does it match any declared interest topics?
• How directly relevant is it?

Quality Signals:

• For repos: stars, recent activity, maintainer health
• For tools: pricing model, maturity, adoption
• For articles: author credibility, publication quality, recency
• For all: uniqueness vs. what's already in the vault

5. Recommendation

Based on analysis, recommend one of two actions:

Save — Worth adding to the knowledge base

✅ SAVE — [Title/Name]
Category: [suggested category for url-dump]
Relevance: [High/Medium] — [why it matters]
Projects: [affected project(s) if any]

Shall I hand off to /url-dump to save it?

Skip — Not relevant or not worth the time

⏭️ SKIP — [Title/Name]
Reason: [clear explanation — wrong stack, low quality, already covered, irrelevant to interests]

6. Batch Summary (for multiple items)

When processing multiple URLs/tools, end with a summary table:

## Scout Summary

| # | Item | Verdict | Reason |
|---|------|---------|--------|
| 1 | [Name 1] | ✅ Save | [brief reason] |
| 2 | [Name 2] | ⏭️ Skip | [brief reason] |

**Actions:**
- [X] items ready to save via /url-dump

7. Execute Follow-up Actions

Based on user confirmation:

• Save items: Hand off to /url-dump with pre-filled category suggestion
• Skip items: No action needed

Fallback Behavior

| Scenario | Behavior |

|----------|----------|

| web-fetch unavailable | Evaluate based on URL structure, domain reputation, and vault search only. Note that content wasn't fetched. |

| web-search unavailable | For tool-name inputs (no URL), ask the user for a direct URL instead. For URL inputs, proceed normally — web-search is not needed. |

| No user profile | Evaluate with general quality/relevance criteria, skip personalized relevance scoring |

| URL is paywalled | Note limitation, evaluate based on available preview and metadata |

| Tool not found via search | Ask user for more context or a direct URL |

Uncertainty Handling

• High confidence: Clear relevance match or clear irrelevance — give direct recommendation
• Medium confidence: Partial match — present pros/cons, let user decide
• Low confidence: Can't determine relevance — explain what's unclear, ask user for context

Integration with Other Skills

Downstream

• Save → hands off to /url-dump with pre-filled category

Upstream

• /daily-brief may surface new tools/services → user can run /scout to evaluate
• /auto-research may discover tools during research → scout can triage them

Success Metrics

• Quick triage (< 1 minute for single URL in solo mode)
• Clear, actionable recommendations
• Accurate vault coverage detection (no duplicate saves)
• Relevance scoring matches user expectations
• Smooth handoff to /url-dump when saving

Philosophy

Scout embodies COG's "evaluate before you accumulate" principle:

• Not everything deserves a bookmark — be selective
• Existing coverage should be surfaced before creating duplicates
• Binary save/skip keeps decisions fast and avoids half-measures
• Clear recommendations reduce decision fatigue