Security for LLM and agent systems: direct and indirect prompt injection, the OWASP Top 10 for LLM Applications, MITRE ATLAS technique mapping, excessive agency and tool-scope containment, egress allowlisting to stop data exfiltration, human confirmation for irreversible actions, and treating third-party skills and MCP servers as untrusted code. Use when an agent is given tools or credentials, when retrieved documents or repository files could carry injected instructions, or when reviewing an AI feature before it reaches production.
Choose the right grounding layer for an agent across the Microsoft IQ stack - Work IQ, Fabric IQ, Foundry IQ, Web IQ. Use when an agent can see someone's email, calendar and Teams messages but not the approval rules or business entities it needs, when deciding between workplace signals and an ontology over business data, when someone maps a conceptual \"Business IQ\" or \"Work IQ\" onto Microsoft's products and gets the opposite of what they expected, when an agent needs GL accounts, vendors or authorisation thresholds rather than documents, or when an unattended background agent is being designed on top of Work IQ.
This rule is essential for maintaining consistency and quality in rule creation across the codebase. It must be followed whenever: (1) A user requests a new rule to be created, (2) An existing rule needs modification, (3) The user asks to remember certain behaviors or patterns, or (4) Future behavior changes are requested. This rule ensures proper organization, clear documentation, and effective rule application by defining standard formats, naming conventions, and content requirements. It's particularly crucial for maintaining the rule hierarchy, ensuring rules are discoverable by the AI, and preserving the effectiveness of the rule-based system. The rule system is fundamental to project consistency, code quality, and automated assistance effectiveness.
119mo agoReviewMountCreate a free account to mount
Generate real starter content for a project's What-L2 (requirements) and How-L2 (architecture/conventions) CEP layers once ult-repo-layout has resolved their paths but found them empty — an honest, minimal, YAML-frontmatter-first overview document per layer for small/single targets, or graphify-informed per-module tiering with resumable per-module CONTEXT.md generation, existence-gated CODING-STANDARDS.md/TESTING-GUIDELINES.md, interface-boundary docs for graph-crossing module pairs, and a rendered CEP-INDEX.md router for large repos, optionally informed by a user-supplied domain-pack of terminology/references if one is configured. This skill generates a NEW starter doc only when none exists — it never compiles or reconciles EXISTING scattered guideline sources (that's compiling-project-guidelines' job). Do NOT use to enforce layout paths or run layer discovery — that's ult-repo-layout. Do NOT use to author or generate a domain pack — this skill only ever consumes one you already wrote.
Explains where different kinds of content belong in this repo (CLAUDE.md vs .claude/skills/ vs docs/), how to build and strict-check the MkDocs Material site, and the writing conventions used throughout (Roadmap admonitions, relative links, ADR format). Use this whenever writing or editing any file under docs/, whenever deciding whether new content belongs in CLAUDE.md, a skill, or docs/, whenever the user asks about documentation structure, or before adding a page to mkdocs.yml's nav. Make sure to consult this before adding procedural, multi-step content to CLAUDE.md directly — CLAUDE.md is deliberately kept short and holds only facts; procedures belong in a skill instead, and getting this wrong bloats the context every single session pays for.
Explains claudeloop's testing philosophy and pytest layout — fakes over mocks for every port, FakeClock/FakeSleeper for simulating multi-day waits without real sleeping, mandatory Hypothesis property tests for numeric/time-based invariants, per-layer coverage gates (100% for domain and application), and the documented # pragma no cover policy. Use this whenever writing or modifying any test under tests/, whenever the user asks how to test something in this codebase, mentions coverage, Hypothesis, property-based testing, or asks why a coverage gate failed. Make sure to consult this before adding a mock (this codebase uses fakes implementing real Protocols instead), before adding a numeric or time-based config field without a property test, and before adding a # pragma no cover without a stated reason — all three are enforced review expectations here, not suggestions.