## Goal Give AI coding agents (and human CLI users) first-class search capabilities through the `glab` CLI and the `glab` AI skill. Today, agents that need to search GitLab must hand-craft `glab api` calls with URL-encoded paths, manual pagination, and raw JSON parsing. This epic adds a native `glab search` subcommand to the CLI and teaches the `glab` AI skill how to use it — making search a composable, discoverable primitive for both humans and agents. ## Scope 1. **`glab search` CLI subcommand** (`gitlab-org/cli`) — a new top-level command that wraps the GitLab Search API (instance, group, and project levels) with human-readable table output and JSON mode. 2. **`glab` skill update** (`gitlab-org/ai/skills`) — add search guidance, examples, and agent-oriented gotchas to the existing `glab` skill so AI agents know how and when to search. ## Dependency graph ``` Issue 1: Design RFC / proposal ↓ blocks Issue 2: API client layer ──────────────┐ ↓ blocks │ Issue 3: Phase 1 scopes (MVP) │ ↓ blocks │ Issue 4: Phase 2 scopes │ ↓ blocks │ Issue 5: Phase 3 scopes │ │ Issue 6: Skill — glab api search ← independent, can start immediately ↓ blocks (after Issue 3 merges) Issue 7: Skill — native glab search section ↓ blocks Issue 8: Skill — search reference doc + agent guidelines ``` **Two parallel tracks:** - **CLI track:** Issues 1 → 2 → 3 → 4 → 5 - **Skill track:** Issue 6 (immediate) → 7 (after Phase 1 merges) → 8 ## Success criteria - `glab search issues "query"`, `glab search blobs "code"`, etc. work at instance, group, and project levels - The `glab` AI skill teaches agents to use search effectively (including scope restrictions, tier limitations, and URL-encoding pitfalls) - Table + JSON output for all scopes