Skip to content

jeffabailey/jbb-skills

v1.7.1Unlicense

Project fitness review skills (architecture, security, reliability, testing, performance, algorithms, data, accessibility, usability, process, maintainability) plus commit, root-cause, AI-tell cleanup, and fitness-config setup helpers.

agent-hygiene

Cleans up the instruction files agents read in a repo (AGENTS.md, CLAUDE.md, GEMINI.md, nested agent files, Cursor and Copilot rules, skills, subagent definitions). Audits every line for no-ops, duplication, stale claims, restated config, negation, and sprawl, then runs three subagents, each proposing a more radical restructuring than the last: prune, progressive disclosure, rebuild. Compares them by always-loaded size and lost instructions, and applies only the one the user picks. Use when the user says /jbb-skills:agent-hygiene, says their AGENTS.md or CLAUDE.md is a mess, bloated, or contradictory, asks to remove no-ops, use progressive disclosure, move rules into CODING_STANDARDS.md, or slim down agent instructions.

ai-sanitize

Removes the tells that make prose, user interfaces, and graphics read as AI-generated, then reports what changed. Rewrites performative phrasing, contrast framing, triads, signposting, emdashes, and marketing filler in prose; strips purple gradients, glassmorphism, pulsing badges, eyebrow labels, pill-everything, emoji decoration, and landing-page treatment of functional screens from UI code; and cleans decorative blobs, misaligned SVGs, broken ASCII art, and rainbow palettes from graphics and diagrams. Use when the user says /jbb-skills:ai-sanitize, asks to "remove AI tells", "de-slop", "make this sound less like AI", "make this UI look less AI-generated", "humanize this copy", or asks why a page or draft feels generated. Edits in place by default; pass "report" to list tells without editing.

ask-why

Investigates a question about a codebase or system and answers it from evidence. Gathers evidence first (git history and blame, CI run logs, tests, ADRs and docs), then uses Five Whys for 'why is X failing / why did X break', a scored placement analysis for 'where should Y live' (module, layer, script, package, CI job, or bounded context), and structured explanation, description, or a decision matrix for how/what/which questions. Checks the question's premise and says what evidence it could not reach. Use when the user says /jbb-skills:ask-why, asks 'why is this failing', 'why did this break', 'what was the root cause', 'where does this belong', 'where should this check/code live', or wants a root cause explained for a bug that is already fixed. Read-only: no code is written or changed.

button-up-documentation

Tidies a repo's documentation. By default collapses the design docs in docs/architecture/ into ADRs in docs/adrs/ (one decision per ADR, in the repo's own ADR format), deletes docs that are stale, duplicated, or leftover working notes, repoints links, and tightens the prose that stays with Strunk & White's rules. A tidy-only mode cleans prose and removes dead docs without touching ADRs. Then offers the same cleanup for README.md, CONTRIBUTING.md, and other docs. A bundled script inventories the docs and checks links; the skill shows a plan and waits for approval before it edits or deletes anything. Use when the user says /jbb-skills:button-up-documentation, asks to button up, clean up, consolidate, prune, or tidy docs, wants architecture or design docs turned into ADRs, asks which docs are stale or redundant, or wants a README or CONTRIBUTING tightened, even if they don't say 'ADR'.

deslop-python

Audits a Python project for junk tests, dead code, and unused dependencies with evidence from pytest (per-test timings), pytest-cov (per-test coverage), mutmut (which mutations the suite kills, and which kills a test removal would lose), ruff, ty, vulture, radon, deptry, jscpd, and import-linter. Installs missing tools, writes a prioritized report (file and test, problem, evidence, recommended optimization, estimated time saved, risk), changes no test or source file until the user approves, then applies only approved items and verifies them against the original test and mutation results. Use when the user says /jbb-skills:deslop-python, wants to clean up, slim down, dedupe, or speed up a pytest suite, asks which Python tests are redundant, weak, slow, shadowed, or safe to delete, wants to parameterize tests or consolidate fixtures, asks for a mutation-testing or coverage-overlap review of a Python project, or wants dead Python code or unused dependencies found and removed, even if they don't name the tools.

deslop-rust

Audits a Rust crate or workspace for junk tests and dead code with evidence from cargo-nextest (per-test timings), cargo-llvm-cov (per-test coverage overlap), and cargo-mutants (which mutations the suite catches, and which kills a test removal would lose). Installs missing tools, writes a prioritized report (file and test, problem, evidence, recommended optimization, estimated time savings, risk), changes nothing until the user approves, then applies only approved items and verifies them against the original test and mutation results. Use when the user says /jbb-skills:deslop-rust, wants to clean up, slim down, dedupe, or speed up a Rust test suite, asks which Rust tests are redundant, weak, slow, or safe to delete, wants to parameterize or consolidate Rust tests, asks for a mutation-testing or coverage-overlap review of a Rust project, or wants dead Rust code found and removed, even if they don't name the tools.

fitness-config-init

Creates a project's fitness-config.json with review weights that fit its purpose. Scans the folder (or runs a full review first, if asked), classifies the purpose, proposes weights from a matching purpose profile, explains every change, and saves only the proposal the user approved. Use when the user says /jbb-skills:fitness-config-init, asks to "create a fitness config", "set up fitness-config.json", "initialize fitness config", "tune review weights for this project", "weight the review for this repo", or wants the fitness review to reflect a project's purpose instead of the defaults.

generate-commit

Writes a conventional commit message for what the user has staged, matching the types and scopes in the repo's own git log, runs the project's CI lint on the staged files only, shows the message for review, and commits only after the user confirms. Respects the user's staging, warns about secret-looking or scratch files, and suggests splitting diffs that mix unrelated changes. Use when the user says /jbb-skills:generate-commit, "write a commit message", "commit what I staged", "commit this", or wants a reviewed conventional commit.

review-accessibility

Reviews frontend code for accessibility and usability fitness, scoring semantic HTML, keyboard navigation, screen reader support, color/contrast, progressive enhancement, responsive design, and usability heuristics. Computes WCAG contrast ratios with a bundled script that resolves CSS custom properties per theme (light/dark). Use when the user says /jbb-skills:review-accessibility, requests an accessibility audit, asks for a11y review, wants WCAG compliance check, asks to check color contrast, alt text, or keyboard navigation in templates, components, or CSS, or needs usability evaluation of frontend code. For walking a live site's tasks in a browser, use review-usability. Only reports findings with confidence >= 7/10.

review-agent-memory

Reads what coding agents learned in a project (Claude Code auto-memory, CLAUDE.md/AGENTS.md/GEMINI.md and Cursor/Copilot rules, session transcripts, and context stores such as CONTEXT.md, dual-graph, CodeGraph, and graphify) and ranks the architecture and harness fixes that cut token usage and agent churn. Mines transcripts for repeated and oversized reads, search loops, failed commands, permission denials, subagent fan-out, and hook-injected context, treats each gotcha memory as a gap the code should close, writes a report, and applies only what the user picks. Works on one repo or a workspace. Use when the user says /jbb-skills:review-agent-memory, asks why agents burn so many tokens or keep going in circles in a repo, wants to reduce token usage, agent churn, or context bloat, asks what the agent's memory says about the codebase, wants to turn memories or lessons learned into code or tooling fixes, or asks how to make a project cheaper or easier for agents to work in, even if they don't say 'memory'.

review-algorithms

Analyzes code for algorithm/data structure correctness, concurrency safety, and edge case handling, producing fitness scores (1-10) across algorithm choice, data structure selection, complexity awareness, concurrency safety, edge case handling, and correctness patterns. Use when the user says /jbb-skills:review-algorithms, requests an algorithm review, asks about correctness of data structure choices, wants a concurrency safety check, asks about edge case coverage, or wants to verify algorithmic complexity. This is NOT a performance review -- it asks "is this correct and appropriate?" not "is this fast enough?" Only reports findings with confidence >= 7/10.

review-apply

Takes a fitness report from a GitHub issue or a local file, fixes its action items in the codebase, and closes the issue when done. Use when the user says "apply review", "fix review issues", "address review feedback", "apply fitness report", "work through the action items", or provides a GitHub issue URL or number containing a fitness report, even if they don't say "fitness". Not for producing a new report (that's review-full).

review-architecture

Analyzes code for software architecture fitness, producing scores (1-10) across coupling, cohesion, layering, modularity, naming, API design, and maintainability. Use when the user says /jbb-skills:review-architecture, requests an architecture review, asks about coupling and cohesion, wants to analyze design or check code structure, asks to review naming or API design, or needs architecture fitness scores. Triggers on "architecture review", "coupling and cohesion", "analyze design", "check code structure", "review naming", "API design". Only reports findings with confidence >= 7/10.

review-data

Analyzes database schema design, migration safety, data integrity, query correctness, data modeling, and pipeline quality, producing fitness scores (1-10) with file:line evidence. Use when the user says /jbb-skills:review-data, requests a data review, asks about schema design, migration safety, data integrity, query correctness, data modeling, or pipeline quality. Distinct from review-performance (which asks "are queries fast?"); this asks "is the schema correct? are migrations safe? will data integrity hold?" Only reports findings with confidence >= 7/10.

review-full

Runs a comprehensive project fitness review combining architecture, security, reliability, testing, performance, algorithms, data, accessibility, process, and maintainability analysis. Use when the user says "full review", "comprehensive review", "project fitness", "review everything", or wants all review skills run on current changes before shipping.

review-jit-test-gen

Generates just-in-time catching tests for changed code. Finds uncovered paths in pending changes and writes focused tests that catch regressions at the moment of change. Use when the user says "generate tests", "jit tests", "test generation", "write tests for changes", or wants tests for recently modified code.

review-maintainability

Analyzes code for maintainability, understandability, and simplicity fitness, producing scores (1-10) across structural complexity, comprehensibility, technical debt indicators, coupling/dependency depth, and code smell density. Use when the user says /jbb-skills:review-maintainability, requests a maintainability review, asks about code complexity or understandability, wants cyclomatic/cognitive complexity analysis, or needs simplicity/maintainability fitness scores. Triggers on "maintainability review", "understandability", "code complexity", "cognitive complexity", "cyclomatic complexity", "simplicity", "code smells". Only reports findings with confidence >= 7/10.

review-performance

Analyzes code for performance and scalability issues, producing fitness scores (1-10) across algorithmic efficiency, database design, caching strategy, scalability readiness, resource utilization, and data pipeline efficiency. Covers services, CLIs, CI scripts, batch jobs, and ML inference code. Use when the user says /jbb-skills:review-performance, requests a performance review, asks for scalability analysis, wants to find N+1 queries or Big-O hot paths, asks why something gets slow as data grows, or needs performance fitness scores before shipping. Only reports findings with confidence >= 7/10.

review-process

Evaluates a repository's development process maturity across documentation, workflow, code review, dependency management, project organization, portability, and leadership signals, using repo files and git history. Use when the user says /jbb-skills:review-process, requests a process review, asks for development process fitness scores, wants to assess repo health or contributor readiness, or asks how well a project follows software development best practices. Only reports findings with confidence >= 7/10.

review-reliability

Analyzes code and configuration (application code, Docker Compose or IaC stacks, and CI-only pipelines) for production reliability fitness, producing scores (1-10) across observability, availability design, timeout/retry hygiene, CI/CD maturity, incident readiness, capacity planning, and container/deploy hygiene. Use when the user says /jbb-skills:review-reliability, requests a reliability review, says check observability, asks if the system is production ready, wants a monitoring setup review, asks about CI/CD pipeline quality, or wants an operational readiness assessment. Only reports findings with confidence >= 7/10.

review-security

Analyzes code, CI/CD workflows, and container config for security vulnerabilities and compliance risks, producing fitness scores (1-10) across input validation, authentication/authorization, data protection, dependency security, error handling/logging, and cryptography. Use when the user says "security review", "check vulnerabilities", "audit security", "security issues", /jbb-skills:review-security, asks to audit GitHub Actions workflows or a Dockerfile for security, or wants security fitness scores before shipping. Only reports findings with confidence >= 7/10.

review-testing

Analyzes a codebase for software testing fitness, producing scores (1-10) across test pyramid balance, test quality, coverage strategy, performance testing, debugging support, and CI integration. Use when the user says /jbb-skills:review-testing, requests a test quality review, asks about testing strategy, wants test pyramid analysis, asks about QA practices, or needs testing fitness scores before shipping. Only reports findings with confidence >= 7/10.

review-usability

Reviews a live website's usability by walking its real user tasks in a browser and scoring learnability, efficiency, memorability, error prevention, and satisfaction. Takes its rubric from the current Fundamentals of Software Usability article on jeffbailey.us, or from a bundled copy when offline. Use when the user says /jbb-skills:review-usability, asks to improve a website's usability, wants a usability audit or task walkthrough of a URL, or asks why visitors struggle to find or finish things on a site. For code-level accessibility or WCAG checks of templates and components, use review-accessibility instead. Only reports findings with confidence >= 7/10.

the-deterministic-maker

Makes agent skills cheaper and more repeatable by moving work from the model to scripts. Measures each skill's load and run cost (tokens and API calls per run, commands the model types by hand, programs it writes at run time) with a bundled script, adds review-agent-memory's memory and transcript evidence, finds steps where the model collects, counts, parses, or checks something a script could do, and reports the baseline and ranked candidates. Then makes the changes the user picks (a tested script plus a shorter SKILL.md step), measures again, and reports before and after with every number marked measured or estimated. Covers every skill in the skills folder, or only those named. Use when the user says /jbb-skills:the-deterministic-maker, asks to make a skill or a skills project more deterministic, wants skills to burn fewer tokens, asks which skills cost the most to run, wants to replace LLM steps with scripts, or asks why a skill does by hand what a script could do, even if they don't say 'deterministic'.

version-skills

Versions agent skills. Raises metadata.version in each changed SKILL.md and the plugin version in plugin.json and .claude-plugin/plugin.json together, using a bundled script, then proves that nothing changed without a version rise. Use when the user says /jbb-skills:version-skills, asks to bump, add, or check skill versions, when a repo's AGENTS.md or CLAUDE.md asks for skill versions to rise, or before committing a change to a repo of skills. For tagging a release of an application, use a release skill instead.