From agent-skills
Persistent, compounding knowledge base for coding agents. Maintains a .wiki/ directory of cross-referenced markdown pages — architecture, bugs, decisions, gotchas, patterns, dependencies. Use when: user says "wiki"/"init wiki"/"ingest"/"lint wiki", starting a long dev session, or the agent discovers something worth persisting (bug root causes, architecture insights, failed approaches). Prevents rediscovering the same things across sessions.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-skills:agent-wikiThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You maintain a persistent knowledge base as plain markdown in `.wiki/` at the project root. The
You maintain a persistent knowledge base as plain markdown in .wiki/ at the project root. The
core insight: the tedious part of a knowledge base is the bookkeeping — cross-references, index
updates, consistency checks. You're good at that. So you own the wiki layer entirely.
The wiki compounds across sessions. Instead of rediscovering that "the auth middleware silently swallows 403s" for the third time, it's already documented with context.
.wiki/pages/ — markdown files you create, update, cross-reference..wiki/SCHEMA.md — conventions governing how you maintain the wiki.Four operations. The user may invoke them explicitly or you may perform them naturally during work.
For concrete input/output examples of each operation, see references/examples.md.
Run when .wiki/ doesn't exist. Sets up the structure and scans the codebase for foundational pages.
Create .wiki/ with SCHEMA.md, index.md, log.md, and pages/ directory.
Use the templates in references/schema-template.md.
Scan the codebase — read the README, manifest files (package.json, Cargo.toml, pyproject.toml, go.mod), and top-level directory structure.
Create 3-5 foundational pages covering: project overview, key architecture, main dependencies.
Update index.md and log.md.
The goal is a useful starting point, not exhaustive documentation. Capture the big picture. Depth comes later through incremental ingests.
The most common operation. When you learn something significant — a bug root cause, an architecture insight, a dependency quirk, a failed approach — file it.
A single discovery often touches multiple pages. Finding a bug might reveal an architecture pattern and a dependency gotcha.
Read index.md to see what exists. Update existing pages or create new ones.
Write pages in .wiki/pages/ with YAML frontmatter:
---
title: Auth Middleware Architecture
type: architecture
created: 2026-04-11
last_updated: 2026-04-11
tags: [auth, middleware, express]
related: [jwt-config, session-handling]
---
Keep pages focused — one topic per page. Split and cross-reference when content spans concerns.
Add [[page-name]] wikilinks. When linking A to B, check if B should link back to A.
Update index.md with new/changed entries under the appropriate category.
Append to log.md with date, operation, summary, and pages touched.
When to ingest without being asked:
The bar: "would another agent waste time rediscovering this?" If yes, ingest it.
When you need codebase context — before starting work, investigating a bug, making a design decision — check the wiki first.
.wiki/index.md to find relevant pages.Querying is lightweight. Don't make it a ceremony.
Run periodically (start of session, or when asked) to keep the wiki healthy.
Check for:
last_updated is old and content may not reflect current code[[wikilinks]] from other pages[[wikilinks]] pointing to nonexistent pagespages/ not in index.md, or vice versaFix what you can, flag what you can't as open questions on the relevant pages.
Compound, don't duplicate. If a page exists, update it. If two pages overlap, merge them.
Focused pages over comprehensive ones. One topic per page. Split and cross-reference.
Write for future agents. Be precise — include file paths, function names, error messages. Skip prose.
Low ceremony. A rough page capturing a key insight beats no page. Lint and improve later.
Cross-reference aggressively. The wiki's value grows with connections. Update related pages too.
.wiki/
├── SCHEMA.md # Conventions and page type definitions
├── index.md # Categorized catalog of all pages
├── log.md # Append-only chronological activity log
└── pages/
├── project-overview.md
├── auth-middleware.md
└── ...
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub pjordan/agent-skills --plugin agent-skills