skills

Shared engine for the decision-trail suite (track-logic, toggle-track-logic, show-logic). Not invoked directly — it provides the storage, config, branch resolution, capture hooks, and read/audit/reconstruct helpers the three commands call. Read this to understand or maintain the runtime.

Use to show the decision trail for the current branch — what was decided, why, with what perceived confidence, and on what evidence — as a reviewable table plus decision-quality notes. Invoke as /show-logic [branch] [--pr]. Collects rows across parallel worktrees, validates evidence against the diff, flags conflicts and problematic choices, and reconstructs a best-effort trail (clearly marked) when tracking was off. Use when the user asks to see or audit the rationale/logic/decisions behind a branch or PR, or to add it to a PR.

Use to record a decision in the branch's decision trail — why a fork was chosen, an approach rejected, a scope call made, and the decision-maker's perceived confidence — by you or on the user's behalf. Owns the row protocol the whole logic suite shares. Invoke as /track-logic "<what> because <why>" [--confidence high|medium|low], and follow its capture rules whenever tracking is on for the branch.

Use to turn the decision trail on or off for a branch. Invoke as /toggle-track-logic [on|off] [branch] — no state flips the current setting, no branch targets the current checkout. Enabling materializes the committed .logic/ runtime and the capture hooks. Use when the user asks to start or stop tracking decisions/rationale/logic for a branch or the repo.

Security audit and active remediation for agent skills. Analyzes SKILL.md instructions and bundled scripts for prompt injection, data exfiltration, excessive permissions, supply chain risks, and other threats. Presents findings inline, optionally generates reports, and can rewrite skills to remove security concerns.

Explore and analyze TUI applications to document their features for cloning. Use when asked to reverse-engineer, analyze, document, or understand a terminal UI like Claude Code, OpenCode, Codex, lazygit, or any ratatui/ncurses-based application. Launches the target TUI in tmux, systematically explores all views and keybindings, captures ASCII diagrams of each screen, and writes findings incrementally to a markdown file (survives context compaction).

Analyzes staged git changes and evaluates agentic context files (Claude, Codex, Cursor, etc.) to suggest additions or removals. Use after staging changes before committing.

Conducts time-boxed technical investigations, competitive research, or feasibility analysis. Asks clarifying questions upfront and produces a structured markdown report with executive summary. Use for technical spikes, architecture decisions, or evaluating new technologies.