useful-ai-agent

Create, verify, restore, and protect encrypted local Git bundle backups without deleting user data.

Maintain markdown files as persistent agent memory with routing, append-only timelines, source-of-truth rules, and low context bloat.

Keep markdown source files and MemPalace memory aligned after new durable facts or compaction.

Operate Nanobot Telegram agent, WebSocket, Guest Mode, effort markers, and health checks safely.

Template for creating a personal writing voice skill from examples without leaking private source material.

Quarterly maintenance audit for an existing context architecture. Checks for entropy: token count drift, orphan files, routing violations, stale content, broken cross-references, and "new agent" test regression. Outputs a cleanup report with proposed actions. Use when: architecture was set up (by context-architecture-setup or manually) and needs maintenance, agent quality has degraded over time, token count has grown, user suspects file sprawl, user wants a quarterly health check, user says "clean up context", "audit my files", "context is getting messy again", "run cleanup". Also use after major project milestones (launch, pivot, fundraise close) when context shifts significantly. This skill assumes an architecture already EXISTS – if no entry point or routing table exists, use context-architecture-setup first.

Set up a lean, navigable context architecture for AI agents in an organization. Guides the agent through assessment, design, implementation, and verification of lifecycle-split file structure, routing rules, and entry points (CLAUDE.md / AGENTS.md). Use when: a team needs their AI agents to stop hallucinating and start making better decisions, the workspace is messy, or the user wants to set up context for Claude Code / Cursor / Copilot / any AI tool. Trigger on: "set up context", "context architecture", "AI enablement", "agent setup", "file structure for AI", "agent keeps losing context", "agent hallucinating", "set up CLAUDE.md", "set up AGENTS.md".

Iterative quality improvement for any AI output — getting to 9.999/10 through structured critique, re-reading source files, deep research, and rewriting. Use this skill when: the user asks to review, critique, or improve any output; the user says the output is weak, shallow, or "scratching the surface"; the user asks to re-read context and find what was missed; the user wants MECE coverage; the user asks for "non-obvious connections" or a "third perspective"; the user signals high stakes ("you have one chance", "don't miss anything", threats to replace with another agent); the user asks to "think harder" or "go deeper". Also trigger for critique of the user's own documents, strategies, or plans. This skill is about extracting the absolute maximum quality through systematic iteration — not settling for "good enough." After delivering improved output, this skill also reconciles source .md files against audit findings — proposing edits the user must explicitly approve or reject before moving on.