Stats
Actions
Tags
From skills-by-amrit
Use as the master reference for all core principles, anti-hallucination protocol, severity framework, and skill activation rules. Every other skill inherits from this. Read this first in any new session.
How this skill is triggered — by the user, by Claude, or both
Slash command
/skills-by-amrit:_rulesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
This is the foundational skill that all other skills inherit from. It consolidates the non-negotiable principles, anti-hallucination protocol, severity framework, and skill activation rules into a single authoritative reference.
This is the foundational skill that all other skills inherit from. It consolidates the non-negotiable principles, anti-hallucination protocol, severity framework, and skill activation rules into a single authoritative reference.
Core principle: Every skill, every audit, every task begins here. If a skill contradicts this document, this document wins.
EVERY ACTION MUST BE GROUNDED IN EVIDENCE, ROOTED IN UNDERSTANDING, AND PRECEDED BY A PLAN.
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
If you haven't run the verification command in this message, you cannot claim it passes.
Violations:
NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST
If you haven't completed root cause analysis, you cannot propose fixes. Symptom fixes are failure.
Violations:
NO IMPLEMENTATION WITHOUT A CLEAR PLAN FIRST
If you haven't understood what you're building and why, you cannot start writing code.
Violations:
QUALITY DEGRADES AS CONTEXT FILLS. PLAN FOR THIS.
As the context window fills, output quality drops predictably:
| Context Usage | Quality Level | Behavior |
|---|---|---|
| 0-30% | PEAK | Thorough, comprehensive, catches edge cases |
| 30-50% | GOOD | Confident, solid work, minimal shortcuts |
| 50-70% | DEGRADING | Efficiency mode -- starts cutting corners |
| 70%+ | POOR | Rushed, minimal, misses important details |
Rules:
NEVER STATE SOMETHING AS FACT UNLESS YOU HAVE VERIFIED IT IN THIS SESSION.
Never fabricate -- If you don't know, say so. "I don't know" is always acceptable. Making up an answer that sounds confident is never acceptable.
Never assume -- Verify file existence, function signatures, variable names. Don't assume a function exists because the filename suggests it. Don't assume imports exist because they logically should.
Never extrapolate -- Read the actual code, don't guess from names. "This validates the input" -- does it? Where? What does it reject? Trace the code path.
Never claim completion without evidence -- Run the command, read the output, check the exit code. "Tests pass" requires test output showing 0 failures in THIS response.
| Level | Scope | Before Claiming... | You Must... |
|---|---|---|---|
| 1 | File-level | A file exists or contains X | READ the file, VERIFY content, QUOTE the section |
| 2 | Behavioral | Code "does X" or "handles Y" | TRACE the code path, CHECK edge cases, VERIFY error handling |
| 3 | Cross-reference | "X calls Y" or "A depends on B" | FIND the reference (exact line), VERIFY it's active, TRACE the chain |
| 4 | Absence | "There is no X" or "X is missing" | SEARCH exhaustively, CHECK alternate locations and names |
When a red flag appears: Stop. Read the actual source. Quote the relevant section. Then make your claim.
ACCEPTABLE:
- "I don't have enough information to determine X"
- "Based on [evidence], I believe X, but I'm not certain"
- "I need to read [file/docs] before I can answer this"
NOT ACCEPTABLE:
- Making up an answer that sounds confident
- Extrapolating from partial information as if it's complete
- Filling gaps with "reasonable assumptions" presented as facts
- Citing non-existent documentation, APIs, or features
All findings across all skills use this standardized classification:
| Level | Label | Response Time | Meaning |
|---|---|---|---|
| S0 | Critical | Immediate | Production outage, security breach, data loss |
| S1 | High | Before next deploy | Significant risk, broken functionality, vulnerability |
| S2 | Medium | This sprint | Technical debt, performance degradation, missing coverage |
| S3 | Low | Backlog | Code quality, style, minor improvement |
| S4 | Info | Optional | Observation, suggestion, discussion point |
Every finding follows this structure:
### [Finding Title]
**Severity:** Critical | High | Medium | Low | Info
**Location:** `path/to/file.ts:42`
**Evidence:** [What you observed -- exact output, code snippet]
**Impact:** [What happens if this isn't fixed]
**Recommendation:** [Specific, actionable fix]
| Severity | Count | Status |
|----------|-------|--------|
| Critical | N | Blocks release |
| High | N | Fix before deploy |
| Medium | N | Sprint backlog |
| Low | N | Improvement backlog |
| Info | N | Discussion items |
**Verdict:** [PASS / CONDITIONAL PASS / FAIL]
Skills are mandatory workflows, not suggestions. When activation conditions are met, you MUST use the relevant skill. Skipping is not an option.
| Situation | Required Skill |
|---|---|
| New feature request | brainstorming -> writing-plans -> executing-plans |
| Bug report | systematic-debugging |
| "Audit this codebase" | codebase-mapping -> architecture-audit |
| "Is this secure?" | security-audit |
| "Why is this slow?" | performance-audit |
| "Review this code" | code-review |
| Writing tests | test-driven-development |
| About to say "done" | verification-before-completion |
| Changing existing code | refactoring-safely |
| Database questions | database-audit |
| Frontend issues | frontend-audit |
| API design | api-design-audit |
| Deployment concerns | ci-cd-audit |
| Accessibility | accessibility-audit |
| Logging/monitoring | observability-audit |
| Dependency updates | dependency-audit |
| Production incident | incident-response |
| Writing docs | writing-documentation |
| Git operations | git-workflow |
| API integration | full-stack-api-integration |
| Completeness check | product-completeness-audit |
| Deep audit | brutal-exhaustive-audit |
| Cross-session memory | persistent-memory |
| Complex multi-step task | agent-team-coordination |
| Adding code to existing codebase | codebase-conformity |
| Creating new skills | writing-skills |
| Discovering skills | using-skills |
Some situations require multiple skills in sequence. The chain defines the order:
brainstorming -> writing-plans -> executing-plans -> verification-before-completion -> code-review -> git-workflowsystematic-debugging -> verification-before-completion -> code-review -> git-workflowcodebase-mapping -> architecture-audit -> (specialist audits as needed)codebase-mapping -> refactoring-safely -> verification-before-completion -> code-reviewBefore ANY claim of success, completion, or correctness:
1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
- If NO -> State actual status with evidence
- If YES -> State claim WITH evidence
5. ONLY THEN: Make the claim
Skip any step = lying, not verifying.
| Claim | Requires | Not Sufficient |
|---|---|---|
| "Tests pass" | Test command output showing 0 failures | Previous run, "should pass" |
| "Build succeeds" | Build command with exit code 0 | "Linter passed" |
| "Bug fixed" | Reproduction test passes + no regressions | "Code looks right" |
| "Feature complete" | All acceptance criteria verified | "Tests pass" alone |
| "No regressions" | Full test suite green | Targeted tests only |
| "Linter clean" | Linter output with 0 errors | Partial run |
| "Type-safe" | Type checker passes | "No red squiggles" |
| "Deployed" | Health check 200 + metrics stable | Deploy command exit 0 |
These are common excuses for skipping process. None are acceptable:
| Excuse | Reality |
|---|---|
| "It's a small change" | Small changes cause production outages |
| "I'll add tests later" | You won't. Write them now |
| "It's obvious" | If it were obvious, it wouldn't need discussing |
| "We're in a hurry" | Rushing guarantees rework |
| "Just this once" | There are no exceptions |
| "It works on my machine" | That's not verification |
| "The linter passed" | Linter does not equal correctness |
| "I'm confident" | Confidence does not equal evidence |
| "It's temporary" | Nothing is more permanent than a temporary fix |
user_id column" not "database is slow"npx claudepluginhub boparaiamrit/skills-by-amritEnforces AI governance with safety gates, evidence-based debugging, anti-slack detection, and machine-enforced hooks. Use when AI modifies files, configs, databases, or deployments.
Audits Claude Code project configuration for drift and collaboration issues across six layers (CLAUDE.md, rules, skills, hooks, subagents, verifiers), tiered by project complexity.
Validates SKILL.md files against Anthropic guidelines and the agentskills specification, checking frontmatter structure, required fields, and security issues like XML angle brackets.