Collection of LLM agent skills that I use in my day to day.

Top Skills

• specy: Write flow docs and other useful doc types.
• dev.review: Code review, slop review, spec review, and other review passes.
• sw-loop: Devloop with subagents and loops on review.
• dev.shortcuts: Triggers for workflows like closing and archiving tasks, pushing code, loops, and PRs.
• ag-learn: Automatically improve skills based on conversation sessions.

Drafts

• ag-task: Legacy Markdown task runner retained as a draft while its replacement is pending. Draft skills are not included in runtime sync.
• integ: Legacy standalone integration-harness manager retained as a draft while its replacement is pending. Draft skills are not included in runtime sync.

Conventions

Skills may use these path conventions:

• ROOT_DIR: Base path for skill outputs when a skill writes files (for example research or design docs). Each skill can define a default when this value is not set.
• DOCS_ROOT: Path to docs generated by skills. By default, should be set to the same value as ROOT_DIR

Users can customize either value by declaring it in AGENTS.md.

Example:

## Skill Conventions
- Set `ROOT_DIR` to `./`
- Set `DOCS_ROOT` to `./docs`

Skills

Agent

• ag-dir: Define and maintain an Agent Project Directory (AGD) with clear durable docs, canonical status ownership, and isolated runtime artifacts.
• ag-ledger: Record, sync, and query append-only agent activity ledger entries using local JSONL daily files under META_LEDGER_ROOT (default ~/.llm/ag-ledger), with structured skill metadata and explicit settled, live-frontier, and stable blocked closeout states.
• ag-learn: Improve skills from observed agent friction in sessions, PRs, or audits, ranking evidence-based learning candidates before routing canonical skill edits through $sc.
• agent-browser: Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
• convosearch: Search prior Codex conversations and return up to 3 relevant thread matches with clickable deep links. Use only when explicitly invoked as $convosearch [query] | [filters].
• fast-mode: Operate Codex in fast-mode by skipping skill loading and going directly to task execution, while allowing only explicitly whitelisted skills from a local allow-list file. Use when a user requests fast-mode, minimal overhead, or "no skills unless allowed."
• gen-notifier: Generic desktop notification skill for agents. Send desktop notifications when tasks are complete (or when user input/errors block progress). By default, assume that all jobs will require a notification unless the user says otherwise.
• learning-capture: Extract and consolidate key learnings, insights, and actionable takeaways from the current conversation session. Use when the user wants to capture, summarize, or document what was learned during the chat, create study materials from discussions, or save important discoveries and decisions for future reference. Triggers include requests like "capture learnings," "summarize what we discussed," "create notes from this conversation," "what did I learn today," or "document our key findings."
• mem: Route durable knowledge and read-only project context, infer candidate paths from schema descriptions, and safely materialize managed or explicit repo-owned layouts through one CLI.
• meta.summarize: Summarize Codex activity from ag-ledger using optional scope, lookup, and groupby arguments (meta.summarize [scope] [lookup] [groupby]). Use when asked to summarize the current conversation (convo), summarize sessions that started in the current workspace (workspace), or summarize all sessions across workspaces (all) over current day, last 24 hours, last week, or last month.
• sc: Create, update, or directly optimize skills and SKILL.md content, using discoverable $skill-name references with synchronized dependency metadata plus canonical-source preflight, validation, packaging, and runtime synchronization.
• thread: Fork a same-directory Codex thread by default, or create one without copied parent history when the user explicitly asks for a new/fresh/clean thread. Publish the real child deep link and raw thread ID immediately after prompt handoff, then register and verify the child and, on the name=work profile, create and verify an In Progress Linear issue for team KEV2; do not wait for child-task completion unless explicitly asked.
• secrets: Load and manage local dotenvx credential sets for agent workflows. Use when directly invoked as $secrets or when a task needs credentials from ~/.secrets/.env.*.
• slack-notify: Send explicit Slack notifications through slack-post using Slack credentials loaded by $secrets slack.
• sw-loop: Swarm-based feature delivery workflow that coordinates manager, spec, implementation, review, cleanup, verification, and required checklist tracking across multiple subagents, routing durable artifacts through $mem before choosing paths and turning explicit user proof/completion gates into checklist rows. Use when the user explicitly wants a swarm, parallelized feature work, or a managed loop that should use $sw-ctrl, $specy, $dev.review, and $dev.loop in that order.

Dev

• claw-integ: Run live OpenClaw integration proof against a named claw gateway profile with showboat-v2 evidence, approval preflight gates for existing apps, ffmpeg video proof, and $mem-resolved durable proof storage.
• claw-image-build: Build, publish, and verify OpenClaw Docker/GHCR images, defaulting to bundled Codex, Slack, and diagnostics-otel plugins with multi-arch manifests, GHCR digest proof, offline startup proof, and secret hygiene.
• claw-score: Audit an OpenClaw maturity-scorecard surface into an evidence-backed component score report under the scorecard directory, using required gitcrawl/discrawl evidence, parallel component research, and shared maturity-band Coverage/Quality scoring.
• commits: Write repository-appropriate git commit messages by first checking local history, then matching that repo's conventions. Use when asked to draft a commit message, explain how a repo formats commits, or adapt commit style across different repositories.
• create-task: Create tasks across target platforms. Use when a user asks to create an issue/ticket/task. Currently supports GitHub Issues and always appends a required Context footer with the LLM session to the task body.
• dev.add-license: This skill should be used when the user wants to add a LICENSE file to their repository. Currently supports Apache License 2.0 with automatic copyright information detection from git configuration. Also updates package.json license field if the file exists.
• dev.bootstrap.typescript: Bootstrap or augment TypeScript projects. Use when asked to scaffold a new TypeScript project using the Copier template at /Users/kevinlin/code/copier-typescript-codex, or when asked to add a pre-commit hook that runs formatting and linting in an existing project.
• dev.can-make-public: Scan a repository and its git history for secrets, credentials, private keys, internal URLs, PII, and other sensitive artifacts before making it public. Use when a user asks if a repo is safe to open-source, requests a pre-publication audit, or wants to sanitize a repo for public release.
• dev.code: This skill should be used when performing any coding task including implementing features, fixing bugs, refactoring code, or making any modifications to source code. Provides best practices, security considerations, testing guidelines, and a structured workflow for development tasks.
• dev.codex: “Codex runtime/config management (MCP servers, config.toml, session setup) plus session analysis/summaries. Use only for Codex self‑configuration. Do not use for creating or updating skills or SKILL.md content; use $sc for that.”
• dev.conventional-commits: Explain and apply the Conventional Commits specification for commit messages, including required format, types/scopes, breaking change notation, and examples. Use when asked to draft, validate, or review commit messages or when a repo wants Conventional Commits adoption.
• dev.diagram: Create or update engineering diagrams in plain ASCII or Mermaid, choosing the format that matches the destination medium and surrounding document. Use when an agent needs to add, revise, convert, or simplify architecture, flow, sequence, state, dependency, or decision diagrams in docs, specs, PRs, or Markdown, especially when the user asks for box art, ASCII, or Mermaid.
• dev.do: End-to-end development task intake and execution. Use when the user gives a dev task (feature/bug/refactor) as a file path, pasted description, or git issue and wants it completed; this skill gathers context, asks clarifying questions if needed, then runs the dev.loop workflow to deliver the change.
• dev.llm-session: Derive Codex CLI session IDs and resume interactive sessions. Use when asked to resume Codex sessions, find or map session IDs, inspect ~/.codex/history.jsonl or ~/.codex/sessions, or troubleshoot Codex resume behavior.
• dev.loop: Drive a development task end-to-end from a user-stated goal through planning, context gathering, execution, verification. Use when the user asks to run a devloop, drive a task to completion, or wants a plan-gather-execute-verify workflow with phased commits and CI verification. Also use if user asks to invoke any individual phase of the devloop
• dev.review: Multi-type review skill for code, developer docs, design docs/specs, architecture, UX, and other reviews, including contract and complexity checks for implementation specs. Use when the user asks for a review or critique (code review, docs review, design doc review, spec review, PR review).
• dev.shortcuts: Resolve explicit trigger:<name> workflows, including trigger:close for closing a tracked task before archiving its Codex thread.
• dev.worktrees: Create and inspect Git worktrees, remove already-clean targets conservatively, and transactionally clean one proven-landed worktree and local branch with exact commit identity, dry-run-first execution, resumable journaling, and strict postconditions.
• fin: Finalize a completed task through a selected gh or local landing flow, using a mandatory origin-main gate and a deduplicated heartbeat that resumes deferred cleanup only after live verification of a pending auto-merge.
• linear: Manage Linear issues, projects, comments, and team workflows with the linear-cli terminal client. Use when the user invokes linear, asks to work in Linear from the terminal, or wants issue/project reads or updates performed through linear-cli rather than a plugin or SDK.
• spec-simulate: Simulate implementing a spec against the real source code, then grade the spec as correct, comprehensive, and simple. Use only when explicitly invoked as $spec-simulate or when the user asks to simulate implementation from a spec.
• sudocode: Write consistent sudocode from real code. Use whenever sudocode needs to be written, including flow docs, specs, architecture docs, or code-to-sudocode transformations.

Docs

• docs-audit-v2: Audit documentation rewrites with JSON-first block and line mappings, validation, Markdown reports, and a self-contained HTML viewer.
• docy: Manage reusable reference docs for agent work through a small CLI. Injection and durable AGENTS.md installation discover Markdown files under the bundled references/ tree; aliases remain explicit. Use it for focused guidance including core/main, ref/developer-docs, ref/spec, and the ref/openclaw-docs product overlay.
• explain-diff: Create a self-contained interactive HTML explanation of a code diff, branch, commit, or pull request, including background, intuition, a behavior-oriented walkthrough, and a five-question quiz.
• find-links: This skill should be used when the user wants to fill in TODO links, placeholder links, or missing links in markdown files. Invoke when the user mentions "fill links", "TODO links", "find links", or asks to add appropriate links to concepts in a document.
• hn-title: This skill should be used when the user wants to create, analyze, or improve blog titles for Hacker News submissions. Invoke when the user mentions "HN title", "Hacker News title", wants to optimize their post title, or wants to increase their chances of reaching the HN front page.
• proofread: Proofread posts before publication. Use when asked to review a draft blog post, article, social post, announcement, essay, or other publish-ready writing for spelling, grammar, repeated phrasing, logic or factual mistakes, weak arguments, or empty and placeholder links.
• spec-template: Turn an existing concrete spec into a reusable generic spec template, extracting invariant contracts, replacing implementation-specific details with slots, and preserving the source spec or PR as a worked example.
• specy: Create structured research documentation for codebase exploration and feature investigation, routing durable artifacts through $mem and using $docy ref/spec as the shared writing-style owner for architecture, service-design, feature-design, and feature-spec documents. Specy retains artifact routing, templates, execution planning, lifecycle, and handoff requirements.

Utils

• docx: Comprehensive document creation, editing, and analysis with support for tracked changes, comments, formatting preservation, and text extraction. When Claude needs to work with professional documents (.docx files) for: (1) Creating new documents, (2) Modifying or editing content, (3) Working with tracked changes, (4) Adding comments, or any other document tasks
• imagekit-upload: Upload images to ImageKit from file paths or clipboard, returning the CDN URL for easy sharing and embedding
• pdf: Comprehensive PDF manipulation toolkit for extracting text and tables, creating new PDFs, merging/splitting documents, and handling forms. When Claude needs to fill in a PDF form or programmatically process, generate, or analyze PDF documents at scale.
• xlsx: Comprehensive spreadsheet creation, editing, and analysis with support for formulas, formatting, data analysis, and visualization. When Claude needs to work with spreadsheets (.xlsx, .xlsm, .csv, .tsv, etc) for: (1) Creating new spreadsheets with formulas and formatting, (2) Reading or analyzing data, (3) Modify existing spreadsheets while preserving formulas, (4) Data analysis and visualization in spreadsheets, or (5) Recalculating formulas