oh-my-ai

humaniser

Remove AI-writing tells and rewrite prose to sound natural and human. Triggers: humanise, humanize, de-AI, deAI, sound natural, less robotic, remove AI tells, AI writing, chatbot tone, Wikipedia AI writing signs, voice match, voice calibration. Uses AskUserQuestion for intake; Read/Write for files; WebSearch/WebFetch when the user wants terminology or usage research; Bash when applying edits in a repo. Outputs: interactive Q&A then rewritten text (and optional brief edit log). Do NOT use for: inventing facts, changing the author's stance, weakening safety or compliance language without explicit instruction, or skipping intake when the user's goals and preservation rules are unclear.

agent-init

Creates, updates, or prunes an AGENTS.md for any repository by auditing the codebase, detecting non-discoverable gaps, and drafting minimal high-signal instructions. Triggers: AGENTS.md, agent setup, agent config, init agents, workspace setup, agent instructions, Claude setup. Uses: Read, Bash, Glob, Grep, AskUserQuestion. Outputs: AGENTS.md at repo root (or module-level for monorepos). Do NOT use for: writing README, generating docs, configuring CI, or editing code.

skill-maker

Use this skill any time someone wants to create, scaffold, build, fix, improve, benchmark, or optimize a Tessl/Claude skill — even if they don't say 'tessl' explicitly. If the request involves making a new skill ('create a skill for X', 'build me a skill that does Y', 'scaffold a skill called Z'), fixing or completing an existing one (missing tile.json, broken repo integration, low eval scores, description not triggering), or running and iterating on evals, invoke this skill. The full workflow covers: structured interview → SKILL.md + tile.json + rules/ scaffolding → README/CI repo integration → tessl tile lint → optional Tessl CLI pipeline (skill review, scenario generate/download, eval run) → hand-authored evals or LLM-as-judge fallback → benchmark logging. Do NOT use for: editing application code, debugging, refactoring, writing general documentation, or creating presentations.

llm-wiki

Compiles and maintains a persistent LLM-written markdown wiki between immutable raw sources and answers—the Karpathy LLM Knowledge Base pattern. The agent writes and maintains the wiki; the human curates sources and reads it. Knowledge compounds instead of being re-derived each query. Triggers: llm wiki, persistent wiki, personal knowledge base, wiki maintenance, ingest sources, compound knowledge, index.md, log.md, Obsidian wiki, cross-references, Karpathy wiki pattern, compile wiki, knowledge base. Uses: Read, Glob, Grep, file edits, optional WebSearch, AskUserQuestion when schema or goals are ambiguous. Outputs: updated wiki pages, index, append-only log, citations on query, filed answers, visual outputs. Do NOT use for: mutating raw sources, one-off chat answers with no wiki artifact, or replacing a user-defined wiki schema without reading it first.

reading-synthesis

Rigorous thirteen-part synthesis of a text or talk: deep summary, insights, structure, critique, framework rebuild, and CEO-level takeaways. Triggers: reading synthesis, synthesize this, deep dive, rigorous analysis, deconstruct, book analysis, article analysis, essay breakdown, intellectual synthesis, multi-dimensional analysis, executive summary of ideas, framework extraction. Uses: Read (and related file tools) for attached sources; WebSearch or WebFetch when comparands are missing or context is thin. Outputs: single structured markdown message with fixed section headers per rules/output-sections.md.

discovery-interview

Conduct an interactive discovery interview to produce a structured product specification. Triggers: write a spec, PRD, feature spec, requirements, product requirements, scope a project, brainstorm a feature, flesh out an idea, plan a new project. Uses AskUserQuestion for all user choices; WebSearch/WebFetch when the user wants research. Outputs: user stories, acceptance criteria, technical constraints, prioritized requirements in docs/specs/ per SPEC_TEMPLATE.md. Do NOT use for: implementation, code review, debugging, refactors, or when the user already has a complete spec they only want edited.