fonnesbeck

Triage and act on a review/feedback document, finding-by-finding, with a per-finding response report. Invoke when the user says "address this review", "resolve findings", "triage review", "fix review feedback", or "work through this review". Accepts a review file as argument (e.g. REVIEW.md, IMPLEMENTATION_REVIEW.md, plans/REVIEW-*.md) and produces a per-finding disposition report.

Drive a complete implementation from a plan through a bounded implement → implementation-review → address-review loop, then run an inline simplification pass using the code-simplifier skill instructions; never spawn code-simplifier as a task subagent. Use this skill when the user asks to "run an implementation loop", "implement this plan and iterate until clean", "execute a plan with implementation review", "spawn an implementation reviewer and address findings", or wants an implementation analog of review-loop. Use review-loop instead when the deliverable is only a pre-execution plan or proposal.

Compares a completed implementation against its original project plan to identify gaps, divergences, and opportunities for improvement. Invoke when the user says "compare this to the plan", "check implementation against plan", "audit implementation", "does this match the plan", "review implementation", or "is the implementation complete". Produces a read-only categorized report highlighting what was built, what was missed, what diverged, and where code can be simplified, streamlined, or made more resource-efficient.

Iteratively review and improve a plan until an independent review has no remaining actionable concerns. A bounded wrapper that runs `review-plans` in a fresh independent reviewer subagent each pass and revises the plan with the `address-review` triage pattern. Use this skill when the user asks to "run a review loop", "review and address until clean", "spawn a reviewer and integrate comments", "repeat review until all concerns are resolved", or wants a reusable loop that combines review-plans with address-review. Use implementation-loop instead when the deliverable is executable code, configuration, or documentation produced from a plan.

Rigorous pre-execution audit of software development and data science project plans. Invoke when the user says "review this plan", "audit this proposal", "evaluate this design doc", "review project plan", "check this RFC", "review this data pipeline", "audit this architecture", or "evaluate model design". Produces a categorized markdown report with severity ratings and actionable feedback for the developer who authored the plan.

Interactive Socratic questioning to discover uncertainty before a written plan is ready. Invoke when the user says "stress-test this idea", "find hidden assumptions", "grill me", or asks for Socratic questioning or critical interrogation of their thinking.

Audit a directory of skills for visibility, deterministic vs non-deterministic behavior, and composability. Use when the user wants to review, assess, or improve a skill collection. Produces a structured report with per-skill ratings and cross-cutting findings.

Instructions for a simplification turn injected by pi-skills' code-simplifier extension. Use when code has been edited or written and needs cleanup, or when the user asks to simplify/refactor recently modified code. Do not invoke as a task subagent; apply these instructions inline in the active agent turn, or via the /simplify extension command.

Automatically augment a skill after it produces friction, misunderstandings, or back-and-forth correction. Invoke after a rough interaction with any skill using phrases like "that was painful," "fix that skill," "don't make me explain that again," "that skill misunderstood me," or /skill:lessons-learned. Reads the current session context, identifies the root cause, and surgically edits the target skill's SKILL.md so future invocations incorporate the correction without manual user intervention.

Create a rich, self-contained HTML visualization of an implementation plan. Invoke whenever the user asks to visualize a plan, create a plan overview, make an implementation plan visual, "show me the plan," "plan as HTML," or any request to turn a project plan, roadmap, or architecture into a visual artifact. Also trigger when the user mentions wanting mock-ups, code excerpts, or maximum context about a plan in a single viewable file. Also trigger automatically after a plan is created, saved, or just written (e.g., after `writing-plans` produces a plan file). This skill gathers plan content from conversation or files and produces a styled HTML file with sections for overview, phases, architecture, code/mock-ups, dependencies, risks, and success criteria, then opens it in a browser window using the `browser` tool (requires a graphical environment; falls back to reporting the file path).

Create effective technical presentations with Slidev, focused on data science, statistics, and developer talks. Opinionated toward the neversink theme with the-unnamed as an alternative. Use when the user wants to build, edit, review, or export slide decks. Covers Slidev syntax, neversink/the-unnamed theme features, visual design principles, storytelling structure, and export workflows.

Plan and review work on a daily and weekly cadence using a small prioritized task slate (inspired by the Ivy Lee method) with Todoist time-blocking. Use when the user asks to "plan my day/tomorrow", "time block my tasks", wants a "nightly review", a "weekly review" or "weekly planning" (Fri afternoon through Sun evening), or wants to clean up incomplete tasks from a finished day or week. Provides plan_day (proposes up to six ordered tasks fitted around existing calendar events and writes them back as timed Todoist tasks), nightly_review (reconciles completions, reschedules or un-dates incomplete items), and weekly_review (light retrospective + stalled-project sweep + per-day soft schedule for the coming week via date-only Todoist dates). Depends on the `todoist-api` skill (td CLI) and the `gws-calendar` MCP tools.

Manage dotfiles and secrets with chezmoi. Use for initializing chezmoi, adding/editing configuration files, creating templates for multi-machine setups, encrypting sensitive files, integrating password managers (1Password, Bitwarden, pass, etc.), running scripts during apply, and syncing dotfiles across machines. Triggers on dotfile management, chezmoi commands, configuration templating, or secrets in config files.