avifenesh/agentsys
26 skills1.7k installs24.2k starsGitHub
Install
npx skills add https://github.com/avifenesh/agentsysSkills in this repo
1DebateThis skill provides prompt templates, context-assembly rules, and a synthesis format for structured multi-round debates between two AI tools. One tool takes the proposer role and a different tool takes the challenger role, and both must back claims with specific evidence over one to five rounds. Developers use it to stress-test an idea, play devil's advocate, or surface blind spots before committing to a decision.88installs2Learnlearn researches a topic online and turns the gathered sources into a structured learning guide plus RAG-optimized indexes an agent can later reference. It uses a progressive broad-to-deep query strategy, scores each source on authority, recency, depth, examples and uniqueness, then fetches only the top sources to extract insights. A developer uses it to build a project knowledge base under agent-knowledge/ that Claude reads when answering questions about those topics. Depth is configurable to 10, 20 or 40 sources.84installs3ConsultThis skill runs a cross-tool AI consultation: it queries another AI CLI such as Gemini, Codex, Claude, OpenCode, or Copilot and returns the response. It maps an effort level to a model per tool, can auto-include diff or file context, and supports resuming sessions. Developers use it to get a second opinion or cross-check a decision with a different AI.76installs4Enhance Promptsenhance-prompts analyzes prompts and system prompts for clarity, structure, few-shot examples and output reliability. It flags anti-patterns such as vague instructions, negative-only constraints, aggressive emphasis, redundant chain-of-thought and critical information buried in the middle of the prompt. A developer runs it to tighten a prompt or template before shipping it into an agent. A JavaScript analyzer implements the detection patterns and applies auto-fixes when --fix is passed.69installs5Sync Docssync-docs keeps documentation aligned with code state by combining discovery, analysis, and changelog checks in one workflow. A developer runs it after code changes to find docs referencing changed files, detect version mismatches, removed exports, and outdated examples, and to check the CHANGELOG. It returns structured issues for an orchestrator to act on rather than editing docs itself.69installs6Enhance Agent PromptsThis skill analyzes agent prompt files for prompt-engineering best practices across frontmatter, structure, instruction effectiveness, and tool configuration. It runs 30 pattern checks covering things like the description trigger test, least-privilege tool scoping, and position-aware ordering, and can auto-fix with a --fix flag. Developers use it to improve agent definitions for Claude Code, OpenCode, and Codex.67installs7Perf Profilerperf-profiler runs profiling tools to capture CPU and memory hot paths with evidence. It verifies debug symbols first, captures file:line for each hotspot, and produces a flame graph or equivalent artifact where possible. A developer uses it to pinpoint exactly where time or memory is being spent during a performance scenario. It refuses to profile without a clear scenario and keeps outputs minimal and evidence-backed.66installs8Validate Deliveryvalidate-delivery autonomously checks whether a task is complete and ready to ship. It runs the project's tests and build, verifies extracted requirements are implemented, and compares test counts to catch regressions. It returns a structured JSON verdict with pass/fail per check and specific fix instructions for an orchestrator to retry.66installs9Perf Analyzerperf-analyzer synthesizes the results of a performance investigation (baseline data, experiment results, profiling evidence and tested hypotheses) into a short summary with actionable recommendations. It lists abandoned hypotheses and next steps, and only cites evidence that actually exists in the logs or code. A developer uses it as the final step of a perf investigation to decide whether to continue or stop. If the data is insufficient it says so and requests a re-run.65installs10Perf Benchmarkerperf-benchmarker runs performance benchmarks to establish baselines or validate regressions using strictly sequential runs. It enforces a 60-second minimum run duration (30 seconds only during binary search), a 10-second warmup, and re-running of anomalies, and never runs benchmarks in parallel. A developer uses it to get trustworthy latency numbers before and after a change. It emits a structured JSON metrics block between PERF_METRICS_START and PERF_METRICS_END markers.65installs11DeslopThis skill cleans AI slop from code by scanning for debug statements, ghost code, orphaned infrastructure, and unused exports. It runs a detection script, optionally enhances findings with AST-based repo-map analysis, and ranks results by HIGH, MEDIUM, or LOW certainty. It returns structured JSON in report or apply mode so an orchestrator can auto-fix high-certainty items.64installs12Enhance Skillsenhance-skills analyzes SKILL.md definitions for trigger quality, frontmatter correctness and discoverability. It validates required frontmatter fields, checks that descriptions include trigger phrases like 'Use when', flags unrestricted Bash and unprotected side-effect skills, and grades findings by certainty. A skill author runs it to make a skill auto-discoverable and safe before publishing. With --fix it applies HIGH-certainty auto-fixes.64installs13Orchestrate Revieworchestrate-review runs a multi-pass code review by spawning parallel Task agents, one per review pass, aggregating and deduplicating their findings, then iterating until the code is clean. It always runs code-quality, security, performance and test-coverage passes and adds conditional specialists (database, architecture, API, frontend, backend, devops) based on signals detected in the changed files. A developer invokes it for a thorough or multi-pass review, including the workflow's Phase 9 review loop. Findings are grouped by severity from critical to low.64installs14Perf Investigation Loggerperf-investigation-logger appends structured notes to a performance investigation log, one file per investigation id. Each entry records exact verbatim user quotes, a phase summary, the decisions made with their rationale, and evidence pointers to files, metrics and commands. A developer uses it to keep an auditable trail across the phases of a perf investigation. It never paraphrases user quotes and uses AI_STATE_DIR for the state path.64installs15Enhance Cross FileThis skill analyzes cross-file semantic consistency across agents, skills, and workflows. It runs a JavaScript analyzer that finds tools used in a prompt body but not declared in frontmatter, references to nonexistent agents, duplicate instructions, and contradictory ALWAYS-versus-NEVER rules. All findings are MEDIUM certainty and require human review rather than auto-fix.63installs16Enhance DocsThis skill analyzes documentation for readability, structure, and RAG readiness. It validates links and heading hierarchy, estimates tokens and trims verbose phrasing, and recommends chunk sizes and self-contained sections for retrieval. An AI-only mode packs information densely for agent docs, while the default mode balances human and AI readers, with auto-fixes behind a --fix flag.63installs17Perf Theory Gathererperf-theory-gatherer generates performance hypotheses for a specific scenario by reviewing recent git history and identifying the code paths involved. A developer uses it during performance work to produce up to five evidence-backed hypotheses before running any experiments. It only proposes theories and never suggests optimizations, keeping diagnosis separate from fixes.63installs18Perf Theory Testerperf-theory-tester validates performance hypotheses through controlled experiments. A developer uses it to confirm a clean baseline, apply one change tied to a hypothesis, run repeated validation passes, then revert before the next test. It records metric deltas and an accept, reject, or inconclusive verdict for each experiment.63installs19Discover TasksThis skill discovers tasks from configured sources such as GitHub issues, GitHub Projects, GitLab, and local markdown files, validates them, and presents a ranked list for the user to pick from. It excludes already-claimed tasks and issues that already have an open PR. Developers use it to decide what to work on next or to list open issues across sources.62installs20Drift AnalysisThis skill analyzes project state to detect drift between documented plans and the actual implementation. It classifies plan drift, documentation drift, issue drift, and scope drift using confidence-graded signals, then scores and buckets items into a prioritized reconstruction plan. Developers use it for a reality check comparing docs, roadmaps, and issues against the code.62installs21Perf Baseline Managerperf-baseline-manager manages performance baseline storage and comparison across versions. It enforces a single baseline JSON per version, stored under a fixed perf/baselines path, recording metrics together with environment metadata. A developer uses it to keep per-version baselines consistent so regressions can be detected reliably. It overwrites an older baseline for the same version rather than creating duplicate files.62installs22Perf Code Pathsperf-code-paths identifies the likely implementation paths for a performance scenario before profiling starts. It uses repo intelligence or grep to find entrypoints and handlers, then lists the top candidate files and symbols tied to the scenario with a short reason for each. A developer uses it to narrow where to profile so time is not wasted. It focuses on Rust, Java, JS/TS, Go and Python and keeps to the most relevant 10 to 15 files.62installs23Enhance Orchestratorenhance-orchestrator coordinates the /enhance command's analyzers, running each enhancer in parallel over whatever content exists in a target path. It discovers plugins, agents, CLAUDE.md files, docs, prompts, hooks and skills, spawns the matching enhancer subagent for each, then aggregates the findings into a single report ordered HIGH to LOW certainty. A developer uses it to audit an agent/plugin repo in one pass instead of running each analyzer by hand. It never auto-fixes unless the --apply flag is set.61installs24Enhance HooksThis skill analyzes Claude Code hook definitions and scripts for safety, correctness, and best practices. It classifies the hook type and event, checks frontmatter, matchers, timeouts, and exit codes against a bundled hook knowledge reference, and can apply auto-fixes with a --fix flag. Developers use it to review hooks before they run untrusted or unsafe automation in a session.60installs25Enhance Pluginsenhance-plugins analyzes Claude Code plugin structures, their MCP tool schemas and security patterns against function-calling best practices. It checks plugin.json validity, tool descriptions and parameter schemas, flags risks like unrestricted Bash, command injection and hardcoded secrets, and grades each finding HIGH, MEDIUM or LOW certainty. A developer runs it to review a plugin before publishing and, with --fix, auto-applies HIGH-certainty schema fixes. It applies the 'Intern Test' to tool descriptions.60installs26Repo Intelrepo-intel runs unified static analysis over a repository, combining git-history intelligence, AST symbol mapping, and project metadata via an agent-analyzer. A developer runs it to initialize or incrementally update a repo map, query hotspots and ownership, and check for stale data. It writes repo-intel and repo-map JSON into the agent platform's state directory for other skills to consume.41installs