The Knowledge Registry for AI

Skills give agents know-how.
Specs give them collective engineering experience.

A spec is a versioned markdown document that captures how to build something well — patterns, constraints, examples, edge cases. Unlike docs written for humans, specs are structured for agents: machine-readable TOML frontmatter, stable URLs, signed authorship, and explicit license terms. Think npm, but for engineering knowledge.

Early access42 specs8 namespacesMIT, Apache-2.0, specbird-attribution licenses
  • Skillsknow-howinvokedProcedural know-how an agent invokes to perform a task. Active, on-demand.
  • Rulesconstraintsalways-onConstraints that shape every response without being called explicitly. Encode guardrails and project conventions.
  • SpecsexperienceconsultedEngineering experience. A spec describes how to build a system correctly. Agents consult specs before writing code, like reading a design document.

How it works

  1. 1Install CLI
    # Universal (macOS + Linux) — Recommended
curl -fsSL https://get.specbird.sh | sh

# macOS
brew install specbird

# Node
npm install -g specbird
  2. 2Require a specspecbird install @attia/proxy
  3. 3Agent uses it

    Agent reads docs/ and builds with proven patterns.

Featured Specs

  • @attia/byte-exact-scannerv1.2.0

    Byte-exact credential leak detection using Aho-Corasick multi-pattern matching

    • security
    • credentials
    • scanning
    ↓ 2.4kMIT
  • @common/aho-corasickv2.1.0

    Fast multi-pattern string matching for large-scale document scanning

    • algorithms
    • string-matching
    • performance
    ↓ 8.7kApache-2.0
  • @security/http-proxy-patternsv1.5.2

    Known patterns for HTTP proxy misconfiguration and credential exposure

    • security
    • proxy
    • patterns
    ↓ 3.1kspecbird-attribution⚑ no-trainThis spec's author has opted out of AI training use.
  • @common/encoding-variantsv1.1.0

    Encoding variant normalization for Unicode and byte sequence analysis

    • encoding
    • unicode
    • normalization
    ↓ 5.2kMIT

Agent-ready

  • registry.specbird.com/llms.txt
  • Content-Type: text/spec
  • Every response tells you what to do next.

One command. Agent-ready context.

specbird
$ specbird install @attia/byte-exact-scanner
✓ Resolved @attia/byte-exact-scanner@1.2.0
✓ Fetched 3 dependencies
✓ Wrote docs/byte-exact-scanner.md
→ Agent context ready. 1,847 tokens.

Why specs, not just rules

CLAUDE.md / .cursorrulesSkill librariesSpecbird specs
Versioned❌✅✅
Attributed authorship❌partial✅
License terms❌❌✅
[next] agent actions❌❌✅
Citable in code annotations❌❌✅

For Agents

Specbird speaks directly to agent crawlers. No JavaScript required for spec content — every spec is server-rendered HTML and available as text/spec.

llms.txt
registry.specbird.com/llms.txt
Content-Type
text/spec (default)
Every response
includes a [next] action
Example text/spec response:
[spec]
name    = "@attia/byte-exact-scanner"
version = "1.2.0"
license = "MIT"

[next]
action  = "annotate"
command = "specbird annotate @attia/byte-exact-scanner@1.2.0 \"summary\""