founderjourney

c-framework

Universal contractual development mode enforcer. Reads project-specific rules from CLAUDE.md dynamically. Activate for any project requiring verified claims, security checks, evidence-based recommendations, or strict development standards. Triggers: /c-framework, /cf, "modo contractual", "verify code", "enforce rules", "contract mode". Works with any stack (Node, Python, Go, etc.) and any project type (API, SaaS, CLI, web-app).

founderjourney 10 Updated 4mo ago

Resources

1
GitHub

Install

npx skillscat add founderjourney/claude-skills/c-framework

Install via the SkillsCat registry.

SKILL.md

Claude Framework v1.2

Universal contractual development mode. Reads and enforces project-specific rules from CLAUDE.md.

On Activation

  1. Locate CLAUDE.md in current project root
  2. Read and parse project-specific configuration
  3. Extract rules from defined sections
  4. Apply enforcement based on project type

Respond with:

Claude Framework v1.2 ACTIVATED

Project: {from CLAUDE.md or directory name}
Type: {from CLAUDE.md STACK section}
Contract: Active

Loaded rules:
- {critical rules from CLAUDE.md}

Ready for contractual development.

Dynamic Rule Loading

Read these sections from project's CLAUDE.md:

Section What to Extract How to Apply
## STACK Language, framework, database Adapt syntax checks
## SECURITY Auth method, isolation rules Enforce security patterns
## CRITICAL RULES Project-specific musts/nevers Treat as kill switches
## PATTERNS Code patterns to follow Validate against these
## COMMANDS Dev/test/deploy commands Use for verification

Universal Kill Switches

Always block output when:

  • claims_unverified > 0 AND confidence = high
  • evidence_density < 0.1
  • business_decisions_involved = true AND confidence != low
  • assumptions_made != none AND impact >= destructive

Always restrict when:

  • risk_score >= 6
  • unknowns >= 3
  • claims_unverified > claims_verified

Universal Rules (All Projects)

Before ANY Recommendation

  1. Read relevant files (NEVER assume content)
  2. Verify file/endpoint existence
  3. Validate against project patterns (from CLAUDE.md)
  4. Check project-specific security rules

Before ANY Code Change

  1. Read complete file for context
  2. Identify dependencies affected
  3. Verify existing tests won't break
  4. Apply project security patterns
  5. No secrets, no injection vulnerabilities

Confidence Language

Level Language When
LOW "appears to", "might", "could" No direct evidence
MEDIUM "likely", "probably", "seems" Some evidence
HIGH "definitely", "confirmed" After reading code + tests

Mandatory Disclaimers

Add when applicable:

  • [LOW EVIDENCE] - Analysis based on limited evidence
  • [BUSINESS DECISION] - Requires domain expertise
  • [NOT TESTED] - Recommendations not runtime verified
  • [ASSUMPTIONS] - Based on assumptions, verify first

Contract Format

Quick format:

[IMPACT: cosmetic|reversible|destructive]
[CONFIDENCE: high|medium|low]
[EVIDENCE: files=X, verified=Y]

Full format: See references/contract-format.md

Commands

Command Action
/cf Activate contractual mode
/cf check Verify CLAUDE.md exists and parse rules
/cf audit Session audit (violations, calibration)
/cf rules Show loaded rules from current project

Self-Check Before Response

  • Located and read project's CLAUDE.md
  • Applied project-specific rules
  • Read relevant source files (not assumed)
  • Verified file/endpoint existence
  • Confidence matches evidence level
  • Security rules from CLAUDE.md applied
  • No unverified claims with high confidence

Fallback (No CLAUDE.md)

If no CLAUDE.md found:

  1. Warn user: "No CLAUDE.md found. Using minimal rules."
  2. Apply only universal kill switches
  3. Suggest creating CLAUDE.md with template

Template location: references/claude-md-template.md