Feature Request: Integrate Ambits Code Coverage Tracking into `/sc:analyze` and Coding Skills

[joshLong145]

Motivation

SuperClaude's coding skills currently operate without visibility into how much of a codebase Claude has actually seen during a session. When /sc:analyze assesses code quality, or /sc:implement modifies files, there's no signal telling the user (or the skill itself) whether Claude has read the relevant code or is working with blind spots.

Ambits:

1. Parsing Claude Code session logs (.jsonl files from ~/.claude/projects/)
2. Building a project symbol tree via Serena's LSP cache (.serena/cache/<language>/*.pkl)
3. Producing a per-file and per-symbol coverage report showing Seen% and Full%

Proposal

1. Enhance /sc:analyze with a --focus coverage domain

Add coverage as a new --focus option alongside quality|security|performance|architecture:

/sc:analyze [target] [--focus coverage] [--depth quick|deep]

Behavioral flow when --focus coverage:

1. Run ambits -p . --serena --coverage
2. Parse the text output into structured data
3. Flag files below the configurable threshold
4. Present coverage report with interpretation guidance (high Seen/low Full = "scanned but not deeply read")

When --focus is NOT coverage (e.g., quality, security), append a "Coverage Context" section to the report showing whether the files being analyzed have actually been read. This gives users confidence that the analysis is grounded in what Claude has seen.

2. Coverage awareness in other coding skills

Each skill runs ambits independently (no shared cache between skills):

Skill	Integration	Behavior
/sc:analyze	--focus coverage domain + context section in all modes	Full coverage report or per-file context
/sc:implement	Pre-modification check	Warn when target file has 0% Full coverage; suggest reading first
/sc:test	Post-test cross-reference	Compare test coverage with session coverage to find untested blind spots
/sc:reflect	Session thoroughness metric	Include overall Seen% and Full% in reflection output
/sc:cleanup	Safety check	Warn before deleting code in files with < 50% coverage

3. Per-project configuration

Add coverage thresholds to project-level configuration (e.g., .superclaude/config.json or CLAUDE.md conventions):

{
  "coverage": {
    "warn_threshold": 50,
    "block_threshold": 0
  }
}

• warn_threshold: Seen% below this triggers a warning in skill output
• block_threshold: Full% at or below this triggers a "read first" suggestion in /sc:implement

4. Binary distribution

Open question for maintainers: What's the best option for distributing the ambits binary?

Changes to /sc:analyze Command File

The analyze.md frontmatter would change:

---
name: analyze
description: "Comprehensive code analysis across quality, security, performance, architecture, and coverage domains"
category: utility
complexity: basic
mcp-servers: [serena]
personas: []
---

Usage line updated:

/sc:analyze [target] [--focus quality|security|performance|architecture|coverage] [--depth quick|deep] [--format text|json|report]

New trigger added:

- Code coverage analysis and session visibility assessment needs

Tool Coordination additions:

- **Bash**: ambits CLI execution for coverage analysis
- **Serena MCP**: Symbol tree resolution via LSP cache (sole parsing backend)

Usage Examples

# Full coverage report for the project
/sc:analyze --focus coverage

# Deep coverage analysis with strict threshold
/sc:analyze --focus coverage --depth deep

# Security analysis with coverage context
/sc:analyze src/auth --focus security
# → Report includes: "Coverage Context: src/auth/login.py 100% seen, 85% full"
# → Or warns: "Coverage Context: src/auth/oauth.py 0% seen — consider reading before trusting analysis"

# Quality check — coverage section appended automatically
/sc:analyze src/parser --focus quality
# → Findings section
# → Recommendations section  
# → Coverage Context section (auto-appended)

Technical Details

How ambits maps tool calls to coverage:

Tool	ReadDepth	Meaning
mcp__acp__Read, Read, Edit	FullBody	Claude read/edited the full file
mcp__serena__find_symbol (with body)	FullBody	Claude read a specific symbol's implementation
mcp__serena__find_symbol (no body)	Signature	Claude saw the symbol signature only
mcp__serena__get_symbols_overview	Overview	Claude saw the file's symbol structure
Grep, mcp__serena__search_for_pattern	Overview	Claude searched the file
Glob, mcp__serena__find_file	NameOnly	Claude knows the file exists

Backward Compatibility

• All existing /sc:analyze behavior unchanged — coverage is a new opt-in focus domain
• Coverage context sections in other focus modes are informational only — never block execution
• Skills without ambits installed gracefully skip coverage sections (no errors, just a note)
• No changes to SuperClaude's core framework — this is purely additive

Why This Matters

Today, SuperClaude skills trust that Claude has sufficient context to make good decisions — but there's no way to verify that. A developer asking /sc:analyze --focus security on a 200-file project has no idea whether Claude actually read the authentication module or just skimmed its structure.

Ambits closes that gap. By surfacing what Claude has and hasn't seen, skills can make better decisions, users can trust the output more, and the framework moves from "Claude analyzed your code" to "Claude analyzed 87% of your code — here's what it missed."

This is a small addition to the skill system with an outsized impact on developer confidence.

[joshLong145]

Checking in to see if there is any interest. if so i can update the feature request with some more details regarding relevant new features regarding sub agents. I am willing to do the implementation, but want to confirm interest before moving forward.