Scope Check Skill

Purpose: Analyze a prompt and return target repositories + reference repositories, or audit-specific files and repos.

Input: prompt (string) - The user's request; mode (string, optional) - "default" or "audit"

Output:

• Default mode: { targets: string[], references: string[] } - Structured object with repo paths
• Audit mode: { repos: string[], files: string[] } - Repos to audit and specific files to check

---

How to Execute

This is a TEXT ANALYSIS task - analyze the prompt text as your sole input:

Default Mode (targets & references)

1. Scan the prompt for reference indicators (repos mentioned for pattern learning)
2. Scan the prompt for target indicators (repos where changes should happen)
3. Apply default: CWD is target if no explicit target mentioned
4. Return: scope: { targets: [...], references: [...] }
5. Complete in under 200 tokens

Expected output format:

scope: { targets: ["{CODE_ROOT}/metasaver-com"], references: ["{CODE_ROOT}/rugby-crm"] }

Audit Mode (repos & files)

1. Detect repo targets using Step 2 rules (repositories to audit)
2. Detect specific files using Step 2A rules (config files to check)
3. If prompt mentions "audit X" pattern, detect files for X
4. Return: scope: { repos: [...], files: [...] }
5. Complete in under 200 tokens

Expected audit output format:

scope: { repos: ["{CODE_ROOT}/metasaver-com"], files: ["eslint.config.js", "turbo.json"] }

Path Resolution: Replace {CODE_ROOT} with the actual code directory (e.g., /home/user/code/). The calling agent provides the resolved paths.

---

Step 1: Detect Reference Repositories

Reference indicators - repos mentioned for learning/copying, NOT for changes:

Pattern	Example
look at {repo}, check {repo}	"look at rugby-crm for patterns"
similar to {repo}, like in {repo}	"similar to how resume-builder does it"
follow pattern from {repo}	"follow the pattern from multi-mono"
reference {repo}, based on {repo}	"reference the rugby-crm implementation"
how {repo} does it, copy from {repo}	"see how metasaver-com handles this"
{repo} has examples, {repo} shows how	"rugby-crm has several pages like this"

Key distinction: Reference repos are mentioned WITH context clues indicating they're for learning, not changing.

---

Step 2: Detect Target Repositories

Target indicators - repos where changes WILL be made:

Pattern	Example
in {repo}, to {repo}, for {repo}	"add feature to metasaver-com"
update {repo}, fix {repo}, change {repo}	"fix the bug in resume-builder"
create in {repo}, build for {repo}	"create new component in rugby-crm"
{repo}'s {thing}, the {repo} {thing}	"the metasaver-com database"
Direct path mentioned	"/home/user/code/resume-builder/src/..."
Explicit: make changes in, modify, implement	"implement auth in metasaver-com"

Default target: If no explicit target mentioned, use current working directory (CWD).

---

Step 2A: Detect Audit Files (Audit Mode Only)

When mode: "audit" is specified, detect specific config files mentioned or implied:

Prompt Pattern	Files Detected
audit eslint, eslint config	["eslint.config.js"]
audit docker-compose, docker compose	["docker-compose.yml", "docker-compose.yaml"]
audit turbo, turbo config	["turbo.json"]
audit typescript, audit tsconfig, ts config	["tsconfig.json", "tsconfig.*.json"]
audit prettier, prettier config	["package.json"] (prettier field)
audit vite, vite config	["vite.config.ts"]
audit vitest, vitest config	["vitest.config.ts"]
audit monorepo root, root config	All root-level config files: [".npmrc", "turbo.json", "pnpm-workspace.yaml", "package.json", "tsconfig.json"]
audit all configs, all configuration	All known config files (union of above)
audit package.json, package files	["package.json"]
audit pnpm-lock	["pnpm-lock.yaml"]

File detection logic:

• Match keywords in prompt (case-insensitive)
• If specific file mentioned (e.g., "eslint.config.js"), return exact filename
• If config type mentioned (e.g., "turbo"), return standard filename for that config
• If "all configs" or "monorepo root", return all root-level configs
• Empty files array means no specific file audit requested

---

Step 3: Known Repositories Reference

Repository Name	Type	Keywords
multi-mono	Producer	multi-mono, shared, library, config package
metasaver-com	Consumer	metasaver-com, metasaver.com, main site
resume-builder	Consumer	resume, resume-builder
rugby-crm	Consumer	rugby, rugby-crm, commithub
metasaver-marketplace	Plugin	agent, skill, command, plugin, mcp, claude, marketplace

---

Step 4: Handle Special Cases

Pattern	Targets	References
No repo mentioned at all	[CWD]	[]
Only reference indicators found	[CWD]	[matched]
sync between X and Y, update both X and Y	[X, Y]	[]
all repos, across all, all my metasaver	[all known repos from Step 3]	[]
standardize X based on Y	[X]	[Y]

---

Examples

Example 1: Clear reference vs target

Prompt: "Add Applications screen to metasaver-com, look at rugby-crm for the pattern"
→ Target: "metasaver-com" (explicit target)
→ Reference: "rugby-crm" (look at = reference indicator)
→ Output: scope: { targets: ["{CODE_ROOT}/metasaver-com"], references: ["{CODE_ROOT}/rugby-crm"] }

Example 2: No repo mentioned (use CWD)

Prompt: "Fix the login bug"
→ Target: CWD (no explicit target)
→ Reference: none
→ Output: scope: { targets: ["{CWD}"], references: [] }

Example 3: Only reference mentioned

Prompt: "Check how rugby-crm handles authentication"
→ Target: CWD (no explicit target for changes)
→ Reference: "rugby-crm" (check how = reference indicator)
→ Output: scope: { targets: ["{CWD}"], references: ["{CODE_ROOT}/rugby-crm"] }

Example 4: Multiple targets

Prompt: "Update scope-check skill in metasaver-marketplace and multi-mono"
→ Target: both repos (explicit update targets)
→ Reference: none
→ Output: scope: { targets: ["{CODE_ROOT}/metasaver-marketplace", "{CODE_ROOT}/multi-mono"], references: [] }

Example 5: Target implicit from context

Prompt: "Create a shared Button component"
→ Target: multi-mono (shared = producer repo keyword)
→ Reference: none
→ Output: scope: { targets: ["{CODE_ROOT}/multi-mono"], references: [] }

Example 6: Standardize pattern

Prompt: "Standardize error handling in resume-builder based on metasaver-com patterns"
→ Target: resume-builder (standardize in)
→ Reference: metasaver-com (based on = reference indicator)
→ Output: scope: { targets: ["{CODE_ROOT}/resume-builder"], references: ["{CODE_ROOT}/metasaver-com"] }

Example 7: All repositories

Prompt: "audit all docker-compose files in all my metasaver repos"
→ Target: ALL known repos (detected: "all my metasaver repos")
→ Reference: none
→ Output: scope: { targets: ["{CODE_ROOT}/multi-mono", "{CODE_ROOT}/metasaver-com", "{CODE_ROOT}/resume-builder", "{CODE_ROOT}/rugby-crm", "{CODE_ROOT}/metasaver-marketplace"], references: [] }

Example 8: All repos with different phrasing

Prompt: "standardize eslint config across all repos"
→ Target: ALL known repos (detected: "across all repos")
→ Reference: none
→ Output: scope: { targets: ["{CODE_ROOT}/multi-mono", "{CODE_ROOT}/metasaver-com", "{CODE_ROOT}/resume-builder", "{CODE_ROOT}/rugby-crm", "{CODE_ROOT}/metasaver-marketplace"], references: [] }

Example 9: Audit single file (Audit Mode)

Prompt: "audit eslint config in metasaver-com"
Mode: "audit"
→ Repos: metasaver-com (explicit target)
→ Files: ["eslint.config.js"] (audit eslint = detected)
→ Output: scope: { repos: ["{CODE_ROOT}/metasaver-com"], files: ["eslint.config.js"] }

Example 10: Audit turbo in all repos (Audit Mode)

Prompt: "audit turbo.json in all my metasaver repos"
Mode: "audit"
→ Repos: ALL known repos (detected: "all my metasaver repos")
→ Files: ["turbo.json"] (audit turbo = detected)
→ Output: scope: { repos: ["{CODE_ROOT}/multi-mono", "{CODE_ROOT}/metasaver-com", "{CODE_ROOT}/resume-builder", "{CODE_ROOT}/rugby-crm", "{CODE_ROOT}/metasaver-marketplace"], files: ["turbo.json"] }

Example 11: Audit monorepo root configs (Audit Mode)

Prompt: "audit monorepo root configuration in multi-mono"
Mode: "audit"
→ Repos: multi-mono (explicit target)
→ Files: [".npmrc", "turbo.json", "pnpm-workspace.yaml", "package.json", "tsconfig.json"] (monorepo root = all root configs)
→ Output: scope: { repos: ["{CODE_ROOT}/multi-mono"], files: [".npmrc", "turbo.json", "pnpm-workspace.yaml", "package.json", "tsconfig.json"] }

Example 12: Audit multiple config types (Audit Mode)

Prompt: "audit typescript and prettier configs"
Mode: "audit"
→ Repos: [CWD] (no explicit target, use current working directory)
→ Files: ["tsconfig.json", "tsconfig.*.json", "package.json"] (combined from typescript and prettier patterns)
→ Output: scope: { repos: ["{CWD}"], files: ["tsconfig.json", "tsconfig.*.json", "package.json"] }

---

Integration

Runs in Phase 1 (Analysis) parallel with agent-check (when used in /audit context).

Output Routing

Default Mode Output (targets + references):

• Passed to: /build, /ms, and general workflow agents
• targets - Repos where workers will make changes
• references - Repos for pattern research (read-only exploration)

Audit Mode Output (repos + files):

• Passed to: /audit workflow and agent-check skill
• repos - Repositories containing files to audit
• files - Specific config files to audit in those repos
• Enables targeted configuration audits without full repository scans