octocode

Use when: applying Matt Pocock engineering skills for diagnosis, TDD, architecture improvement, issue breakdown, PRD creation, or rigorous TypeScript/codebase analysis.

Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.

Sets up an `## Agent skills` block in AGENTS.md/CLAUDE.md and `docs/agents/` so the engineering skills know this repo's issue tracker (GitHub or local markdown), triage label vocabulary, and domain doc layout. Run before first use of `to-issues`, `to-prd`, `triage`, `diagnose`, `tdd`, `improve-codebase-architecture`, or `zoom-out` — or if those skills appear to be missing context about the issue tracker, triage labels, or domain docs.

Break a plan, spec, or PRD into independently-grabbable issues on the project issue tracker using tracer-bullet vertical slices. Use when user wants to convert a plan into issues, create implementation tickets, or break down work into issues.

Turn the current conversation context into a PRD and publish it to the project issue tracker. Use when user wants to create a PRD from the current context.

Use when: working with OpenAI Codex plugins, plugin marketplaces, plugin manifests, plugin-bundled skills, or the openai/plugins repository. Helps inspect .codex-plugin/plugin.json, .agents/plugins/marketplace.json, and installed Codex plugin cache.

Use when: extracting page information, scraping static HTML, comparing rendered page text, using Scrapling, using Skyvern, or deciding between lightweight scraping and browser automation.

Use when: filtering prompts, logs, screenshots, transcripts, documents, or tool outputs for secrets, credentials, PII, private URLs, tokens, keys, or sensitive personal data before sharing with another agent or external service.

Use when: coordinating Codex/Symphony style multi-agent software work, issue-to-PR workflows, review/merge state machines, or the openai/symphony repository.

Periodic self-review skill for Octocode. Surfaces failure patterns, blocked tool calls, slow providers, and circuit-breaker activity from the last operating window so the agent can propose targeted improvements (config tuning, new skills, parser alias additions) and keep the self-iteration flywheel turning.

Browser Harness is a self-healing browser-automation harness from browser-use that lets LLMs complete arbitrary web tasks. Registered as an external reference for OctoCode's exploratory web-automation pathway.

GitNexus is a zero-server, client-side code intelligence engine that builds a knowledge graph of a repository directly in the browser. Registered as an external reference to inform OctoCode's code-intel and graph-aware tool design.

Zed editor's Assistant Panel v2 (right-side dockable AI panel with thread history, inline tool-call disclosure, and slash-command palette). Registered as a UX reference for OctoCode's Canvas WebUI shell.

Routes Claude-Code–style coding agents through NVIDIA's free OpenAI-compatible gateway. In OctoCode this is wired as the `nvidia-free` system fallback provider — used when the operator's configured primary provider is unreachable.

Vue-ecosystem component library. Registered as a UI reference for `ui-shell/` — OctoCode's WebUI is vanilla HTML + IIFE JS today, but Naive UI patterns inform component boundaries for the ongoing modularization.

Autonomous penetration-testing AI agents. Registered as an external integration only; not activated by default. Useful reference for multi-step offensive workflows, sandboxing, and tool gating.

Agent-browser is an LLM-driven browser-automation framework (DOM querying, form filling, navigation) that complements lightpanda-headless-browser for heavier semantic interactions.

Bytebot is an open-source desktop-agent framework exposing OS-level primitives (keyboard, mouse, shell, file I/O) to LLM agents. Registered as reference for uplifting OctoCode's agent tool-calling surface.

LLM-wiki is a curated, living knowledge base about LLM systems programming, prompt engineering, and long-term memory strategies. Registered as a reference skill to lift OctoCode's systems-programming and memory design.

Openharness is an open-source harness for structured LLM evaluation and capability benchmarking. Registered as an external reference to inform OctoCode's evaluation-gate design (see docs/eval/*) and the benchmark gate in runtime/src/benchmarks.rs.

Meta-skill: how to write a SKILL.md that the OctoCode skills hub will pick up. Source of truth for the descriptor schema.

Karpathy-style "think from first principles, verify with tiny scripts" research-engineer patterns. Reinforces the autoresearch + get-shit-down operating loops.

Auto-research loop: decompose a research question, run multiple probes in parallel, reconcile, cite.

Reference library of Anthropic-style product-design patterns for chat UIs — density, latency affordances, tool-call disclosure, streaming indicators.

Long-running memory layer for Claude-style agents. Reference for OctoCode's own `/memories/` scope design (user / session / repo).

Graph-based code-review assistant. Builds a call/dependency graph of a diff and surfaces reviewer-relevant hotspots. Registered for future integration into OctoCode's review workflow.

Ink-style terminal UI toolkit reference. Registered as a UX study for CLI components; OctoCode's terminal surface already uses crossterm + portable-pty but borrows composition ideas from inkos.

Placeholder descriptor for the user-referenced "ligh" project. Upstream identity is ambiguous (possibly "light", "lighthouse", "ligh-browser"); registered as descriptor-only pending clarification.

Convert office / PDF / HTML / image formats to clean Markdown. Prefer this before summarizing document content or extracting structured requirements from attachments.

Long-form narrative writing assistant patterns — chapter planning, continuity tracking, style lock-in. Useful when OctoCode is driven in "writing mode" rather than coding mode.

Adapter layer that fronts heterogeneous model providers behind a single OpenAI-compatible HTTP/1.1 surface so OctoCode's tool-call protocol is real function-calling, not plain-text "<|tool_call|>" markers.

Ultra-light headless browser for fast, low-memory WebUI smoke runs as an alternative to Chromium when a full Playwright stack is too heavy.

Drive the OctoCode WebUI (and other web targets) via Microsoft Playwright to perform real end-to-end verification, screenshots, and DOM assertions.

Protocol-level probe that hits a WebUI over HTTP/SSE/WebSocket and records timing + payload for deterministic health checks.

Single source of truth that lists every tool, MCP server, skill, and hook known to this OctoCode install, and stays in sync with octocode-runtime's tool catalog.

Divergent thinking skill for generating creative ideas, exploring solution spaces, and structured ideation sessions.

Web browser automation skill for navigating pages, filling forms, clicking elements, taking screenshots, and extracting web data using AI-driven browser control.

UI/UX design skill for layout planning, theme creation, animation design, and HTML prototype generation.

Context management skill for selecting, prioritizing, and compressing information sent to LLMs to maximize relevance within token limits.

Personalized learning assistant skill that explains complex topics at adjustable depth, creates study plans, and tests understanding.