tyler-r-kendrick/tkstack
Portable personal learning, quality review, development lifecycle, and measured improvement skills.
Create or maintain source-grounded repository documentation, a knowledge wiki, a glossary, an ADR, or a resumable task handoff. Use for documentation maintenance and knowledge capture. Agent policy belongs in instructify; UI design-system capture belongs in open-design; ordinary articles belong in write.
Run a bounded, evidence-driven improvement loop for code, prompts, skills, websites, or generated media. Invoke only when the user explicitly asks for /hill-climb or a measured improvement loop. Freeze a sensitive evaluation, measure baseline noise, change one hypothesis at a time, keep or revert from evidence, and confirm the winner on untouched cases.
Investigate how a system works, why a decision was made, an unfamiliar API, or the cause of a bug or performance regression. Use for source-grounded explanations, research, diagnosis, and captured-trace analysis. Produces cited findings and explicit unknowns; a diagnosis request alone does not authorize fixes.
Learn from user corrections, frustrations, workflow outcomes, and temporary .tkstack sessions. Use for /learn, help, repeated instructions, a rejected approach, session consolidation, report-only audits, or direct skillify/hookify/instructify/flowify/lintify/reusify/codify subcommands. Preserve provenance, extract scoped reasoning rules, and turn verified procedural lessons into reviewable reusable artifacts shared across harnesses.
Run the BDD/contract lifecycle or scoped planning, architecture, verification, review, delivery, and handoff. Invoke only when the user explicitly asks for /sdlc or one of those modes. Coordinate durable SQLite task DAGs with fenced leases and explicit development-to-requirements/ADR revisions. Full mode retains sparse COW worktrees, independent evidence review, and squash-merged stacked PRs.
Apply tkstack's personal engineering principles when designing, implementing, or reviewing code, skills, hooks, rules, workflows, and harness configuration. Prefer deterministic code and explicit rules, then narrowly scoped decision models or APIs, with LLMs and agents as the last resort. Design all code for safe bounded concurrency and parallel composition.
Identify and correct AI slop in writing, code, tests, websites, images, video, and audio. Use for /unslop, output grading, anti-slop review, or whenever the user identifies a slop pattern. Record actual corrections in the shared steering ledger, extract a scoped reusable criterion, and grade against the evolving catalog with artifact evidence and legitimate exceptions.
Develop a human-facing article, explanation, proposal, or narrative from supplied ideas or source material. Use for drafting, shaping an argument, or a collaborative writing session. Not for ordinary file writes, agent instructions, repository reference maintenance, or anti-slop grading alone.