feat: add list command to oneshot cli

[huberp]

Here's the addendum text to append to the issue:

---

Additional Command: list — Inspect active capabilities

agentloop list <capability> [options]

Capabilities:
  tools           All registered tools (built-in + MCP + custom)
  agentprofiles   All loaded agent profiles (built-in + user-defined)
  skills          All active skills (built-in + user-defined)
  providers       Configured LLM and search providers with their status

Options

      --json      Output as JSON array instead of table
      --verbose   Include full description, permissions, and source path for each entry

Examples

# List all registered tools
agentloop list tools

# List all agent profiles in machine-readable format
agentloop list agentprofiles --json

# List all active skills with full details
agentloop list skills --verbose

# List configured providers and their status (API key present, active, etc.)
agentloop list providers

Sample output: agentloop list tools

NAME              PERMISSION   SOURCE     DESCRIPTION
─────────────────────────────────────────────────────────────────────
code-search       safe         built-in   Search files by literal or regex pattern
file-read         safe         built-in   Read a file from the workspace
file-write        cautious     built-in   Write or overwrite a file
shell-run         dangerous    built-in   Execute a shell command
web-search        cautious     built-in   Search the web via configured provider
web-fetch         cautious     built-in   Fetch a URL and return clean Markdown
plan              safe         built-in   Generate a multi-step execution plan
run               cautious     built-in   Execute a previously generated plan
git-branch        cautious     built-in   Create or list git branches
my-custom-tool    safe         custom     Loaded from TOOLS_DIR
gh-issues         safe         mcp        GitHub MCP server — list/create issues

Sample output: agentloop list agentprofiles

NAME              SOURCE     SKILLS                         DESCRIPTION
──────────────────────────────────────────────────────────────────────────
coder             built-in   typescript, code-review        General-purpose coding agent
planner           built-in   —                              Multi-step task decomposition
reviewer          built-in   code-review, security-audit    Code review and audit agent
devops            built-in   git-workflow                   CI/CD and infrastructure tasks
security-auditor  built-in   security-audit                 Security-focused analysis
my-profile        custom     typescript                     Loaded from AGENT_PROFILES_DIR

Sample output: agentloop list providers

TYPE    NAME          STATUS    NOTE
────────────────────────────────────────────────────────
llm     mistral       ✅ active  MISTRAL_API_KEY set
search  tavily        ✅ active  TAVILY_API_KEY set
search  langsearch    ⚠ inactive  LANGSEARCH_API_KEY not set
search  duckduckgo    ✅ active  no key required
search  searxng       ⚠ inactive  SEARXNG_BASE_URL not set

---

Implementation Notes for list

• Discovery: call ensureInitialized() to hydrate the registries, then read from:
  • toolRegistry.getAll() → tools
  • agentProfileRegistry.getAll() → agent profiles
  • skillRegistry.getAll() → skills
  • appConfig + env var presence checks → providers
• Source tagging: each registry entry already carries metadata; add a source field ("built-in" | "custom" | "mcp") during loadFromDirectory() if not already present.
• Table rendering: use a simple column-aligned padEnd() formatter — no extra dependency. --json outputs JSON.stringify(entries, null, 2).
• Extensibility: list should accept any future capability noun (e.g. agentloop list prompts, agentloop list mcp-servers) — route unknown nouns to a "not yet supported" message rather than erroring hard.

Additional Acceptance Criteria

• agentloop list tools prints a formatted table of all registered tools with name, permission level, source, and description
• agentloop list agentprofiles prints all loaded agent profiles with name, source, active skills, and description
• agentloop list skills prints all active skills with name, source, and description
• agentloop list providers prints LLM and search provider status, indicating whether the required API key / URL is configured
• --json flag outputs a JSON array for all list subcommands
• --verbose flag includes full description, file path, and all metadata fields
• Unknown capability nouns print a usage hint listing supported values and exit 1
• Unit tests cover registry hydration and table/JSON formatting for each capability type