Feat/v0.13 extend coverage

README.md

@@ -6,7 +6,7 @@
 
 A TUI for inspecting agent CLI configurations: what's installed, what's enabled, what's broken.
 
-agentpeek scans your Claude Code config (`~/.claude/` and any project-level `.claude/` directory it finds by walking up from `cwd`) and surfaces every loose file the agent depends on: settings, hooks, slash commands, memory files, MCP servers, installed plugins (with their full contents: manifest, skills, agents, commands, hooks, MCPs), and any health warnings detected during the scan.
+agentpeek scans your Claude Code config (`~/.claude/` and any project-level `.claude/` directory it finds by walking up from `cwd`) and surfaces every loose file the agent depends on: settings, hooks, slash commands, memory files, MCP servers, user/project subagents, path-scoped rules, installed plugins (with their full contents: manifest, skills, agents, commands, hooks, MCPs), and any health warnings detected during the scan.
 
 ## Install
 
@@ -22,12 +22,16 @@ Requires Python ≥ 3.11. Linux and macOS only.
 
 ## What you get
 
-- **Three-zone TUI**: sidebar of 9 categories on the left, items list in the middle, detail pane on the right. Drill-down on every category.
+- **Three-zone TUI**: sidebar of 11 categories on the left, items list in the middle, detail pane on the right. Drill-down on every category.
 - **Multi-scope**: user-scoped (`~/.claude/`) and project-scoped (`.claude/` found by walking up from cwd) configurations shown side by side with `[U]` / `[P]` badges. Health checks compare the two and flag conflicts.
-- **Plugin contents**: every installed plugin is opened up - manifest, skills, agents, slash commands, hooks, MCP servers, and the marketplace's auto-update state. Plugin-contributed commands / hooks / MCPs also merge into the top-level categories with a `[plug:<id>]` provenance tag.
-- **Auto-memory**: Claude Code project memories (under `~/.claude/projects/<encoded>/memory/`) are scanned and labelled with their actual project path (resolved from session logs, not the lossy directory encoding).
-- **Health checks**: orphan hook scripts, missing plugin install paths, conflicting key bindings, scope override conflicts, cross-scope layered memory, hooks referencing dynamic env vars (`${CLAUDE_PROJECT_DIR}`, `$HOME`, …) flagged so you know which won't resolve at scan time, and more. Severity-coloured warning cards in the detail pane.
-- **Write actions**: update, enable/disable, uninstall a plugin or refresh a marketplace without leaving the TUI. Calls shell out to `claude plugin …`; agentpeek never edits Claude state files directly. Re-scan is automatic after every action. Pass `--read-only` to disable.
+- **Subagents**: user and project agents under `agents/*.md` are scanned recursively, with full frontmatter (tools, model, permission mode, memory scope, color, and the rest) shown in the detail card.
+- **Path-scoped rules**: `.claude/rules/*.md` files at user and project scope are listed with their `paths:` globs (or marked "always loaded" when the field is absent), so you can answer "why is Claude doing X here?" by category instead of by guessing.
+- **Plugin contents**: every installed plugin is opened up: manifest, skills, agents, slash commands, hooks, MCP servers, and the marketplace's auto-update state. Plugin-contributed commands / hooks / MCPs also merge into the top-level categories with a `[plug:<id>]` provenance tag.
+- **Memory layer**: `CLAUDE.md`, `CLAUDE.local.md` (project-root personal overrides), and subagent persistent memory directories (`agent-memory/` and `agent-memory-local/`) are all surfaced. CLAUDE.md `@path/to/file` imports are resolved recursively (up to 5 hops, cycles detected) so you see what Claude actually loads.
+- **Auto-memory**: Claude Code project memories under `~/.claude/projects/<encoded>/memory/` are scanned and labelled with their actual project path (resolved from session logs, not the lossy directory encoding).
+- **`~/.claude.json`**: user-scope MCP servers, per-project trust state (allowed tools, enabled / disabled `.mcp.json` servers), and OAuth session presence (display-only - the token value is never read or shown). Folded into the MCP and Settings cards.
+- **Health checks**: orphan hook scripts, missing plugin install paths, conflicting key bindings, scope override conflicts, cross-scope layered memory, hooks referencing dynamic env vars (`${CLAUDE_PROJECT_DIR}`, `$HOME`, ...) flagged so you know which won't resolve at scan time, and more. Severity-coloured warning cards in the detail pane.
+- **Write actions**: update, enable/disable, uninstall a plugin or refresh a marketplace without leaving the TUI. Calls shell out to `claude plugin ...`; agentpeek never edits Claude state files directly. Re-scan is automatic after every action. Pass `--read-only` to disable.
 
 ## Keybindings
 
@@ -63,4 +67,4 @@ Built on Textual: the theme follows `Ctrl+P` (palette) and reflows colours acros
 
 ## License
 
-GPL-3.0-only. See `LICENSE`.
+GPL-3.0-only. See [LICENSE](LICENSE).

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "agentpeek"
-version = "0.12.1"
+version = "0.13.0"
 description = "TUI inspector for agent CLI configuration directories like ~/.claude/."
 readme = "README.md"
 requires-python = ">=3.11"

src/agentpeek/models.py

@@ -43,6 +43,13 @@ class SettingsBundle:
     # surfaced as a list so users can see actual script names without
     # opening the filesystem.
     hooks_dir_files: tuple[str, ...]
+    # `agent` setting: default subagent for sessions in this scope.
+    # Set via `.claude/settings.json` and overridable on the CLI.
+    default_agent: str | None = None
+    # `claudeMdExcludes`: glob patterns excluding specific CLAUDE.md
+    # files from the load order. Surfaced because they answer
+    # "why isn't this CLAUDE.md taking effect?" debugging questions.
+    claude_md_excludes: tuple[str, ...] = ()
 
 
 @dataclass(frozen=True, slots=True)
@@ -174,7 +181,26 @@ class Plugin:
     )
 
 
-MemoryKind = Literal["claude_md", "memory_index", "memory_entry"]
+MemoryKind = Literal[
+    "claude_md",
+    "claude_local_md",
+    "memory_index",
+    "memory_entry",
+    "agent_memory_index",
+    "agent_memory_entry",
+]
+
+
+@dataclass(frozen=True, slots=True)
+class MemoryImport:
+    # Resolved import target encountered while expanding `@path/to/file`
+    # references inside a CLAUDE.md body. `resolved_path` is None when
+    # the import couldn't be resolved (file missing, cycle, depth cap
+    # hit); `reason` carries a short marker for the renderer.
+    raw: str
+    resolved_path: Path | None
+    depth: int
+    reason: str | None = None
 
 
 @dataclass(frozen=True, slots=True)
@@ -184,6 +210,59 @@ class MemoryFile:
     has_frontmatter: bool
     kind: MemoryKind
     project_label: str | None
+    # Auto-memory entries belonging to a specific subagent carry the
+    # agent name resolved from the parent directory; otherwise None.
+    agent_name: str | None = None
+    # Resolved @path imports referenced from this body (recursive, max 5
+    # hops, cycles marked). Empty when the file isn't a CLAUDE.md or
+    # contains no imports.
+    imports: tuple[MemoryImport, ...] = ()
+
+
+@dataclass(frozen=True, slots=True)
+class Agent:
+    # User- or project-scope subagent (`~/.claude/agents/` or
+    # `<project>/.claude/agents/`). Plugin-bundled agents use the
+    # separate PluginAgent type — they live inside Plugin.agents and
+    # are constructed by parsers.plugin_contents.
+    path: Path
+    name: str
+    description: str | None
+    body: str
+    # `tools` and `disallowedTools` accept either a comma-separated
+    # string or a YAML list in the canonical schema. Normalized to a
+    # tuple of token strings here.
+    tools: tuple[str, ...] = ()
+    disallowed_tools: tuple[str, ...] = ()
+    model: str | None = None
+    permission_mode: str | None = None
+    max_turns: int | None = None
+    skills: tuple[str, ...] = ()
+    memory: str | None = None
+    background: bool | None = None
+    effort: str | None = None
+    isolation: str | None = None
+    color: str | None = None
+    initial_prompt: str | None = None
+    # Raw mcpServers + hooks bodies — both have complex schemas that
+    # don't compress into a tuple of strings cleanly. We carry them as
+    # presence flags + a stringified summary so the detail card can
+    # show "configured" / "(none)".
+    has_mcp_servers: bool = False
+    has_hooks: bool = False
+
+
+@dataclass(frozen=True, slots=True)
+class Rule:
+    # `.claude/rules/*.md` — path-scoped or always-loaded context rule.
+    # Per docs, `paths:` frontmatter is optional; absent means the rule
+    # loads at session start with the same priority as `.claude/CLAUDE.md`.
+    path: Path
+    name: str
+    description: str | None
+    paths_globs: tuple[str, ...]
+    always_loaded: bool
+    body: str
 
 
 @dataclass(frozen=True, slots=True)
@@ -214,6 +293,19 @@ class MCPServer:
     auth_pending: bool = False
 
 
+@dataclass(frozen=True, slots=True)
+class TrustEntry:
+    # Per-project trust state from `~/.claude.json[projects][<path>]`.
+    # Surfaced in the Settings detail card so users can see which
+    # projects they've trusted, which tools are pre-approved, and the
+    # per-project enable/disable lists for `.mcp.json` servers.
+    project_path: str
+    trust_accepted: bool
+    allowed_tools: tuple[str, ...]
+    enabled_mcpjson_servers: tuple[str, ...]
+    disabled_mcpjson_servers: tuple[str, ...]
+
+
 @dataclass(frozen=True, slots=True)
 class ScanResult:
     source: str
@@ -226,6 +318,13 @@ class ScanResult:
     keybindings: KeybindingsBundle | None
     mcp: tuple[MCPServer, ...]
     warnings: tuple[ScanWarning, ...]
+    agents: tuple[Agent, ...] = ()
+    rules: tuple[Rule, ...] = ()
+    # `~/.claude.json` highlights, only populated at user scope. Empty
+    # tuple when the file is absent or the scope isn't user.
+    trust_entries: tuple[TrustEntry, ...] = ()
+    oauth_session_present: bool = False
+    claude_json_path: Path | None = None
 
     @classmethod
     def empty(