feat: close agent-host integration parity gaps (CLI fallback, Claude hooks)

CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- Claude Code installs now register `SessionStart` and `PostToolUse` lifecycle hooks, matching the freshness/steering coverage Cursor, Codex, and Kiro already had: `SessionStart` reports index freshness and injects the LCM context-recovery hint after compaction; `PostToolUse` notifies the daemon for targeted incremental sync after edits and shell commands. Existing installs pick the hooks up via the post-upgrade backfill or `tracedecay doctor`.
+- The CLI-fallback steering ("if MCP fails, use `tracedecay tool ...`") now reaches every host with a prompt-rules surface — Claude Code, Copilot/VS Code, Gemini, OpenCode, Kimi, Vibe, and Kiro — instead of only the Cursor rule and Codex session hook.
+
 ### Fixed
 
 - **`serve` no longer exits when project resolution fails at startup** — MCP hosts (Cursor especially) never retry a failed server spawn, so one startup exit over a recoverable config problem (uninitialized project, ambiguous global fallback, bad `--path`) turned every later tool call in the session into "Timed out waiting for connection". `serve` now stays alive in a degraded mode: it completes the MCP handshake, lists the real tools, and answers each tool call with an actionable error naming the failure, the fix, and the `tracedecay tool …` CLI fallback. It rechecks the project on every tool call and recovers in-session once `tracedecay init` (or a corrected path) makes resolution succeed — no server toggle or window reload needed.

docs/USER-GUIDE.md

@@ -184,7 +184,15 @@ TraceDecay works as an MCP (Model Context Protocol) server. AI coding agents con
 tracedecay install
 ```
 
-This is the default. It registers the MCP server in `~/.claude/settings.json`, grants tool permissions so Claude doesn't have to ask you every time, installs a `PreToolUse` hook that redirects Claude away from spawning expensive Explore agents, and adds prompt rules to `~/.claude/CLAUDE.md` that tell Claude to prefer tracedecay tools.
+This is the default. It registers the MCP server in `~/.claude/settings.json`, grants tool permissions so Claude doesn't have to ask you every time, installs lifecycle hooks, and adds prompt rules to `~/.claude/CLAUDE.md` that tell Claude to prefer tracedecay tools (including the CLI fallback for MCP transport failures). The hooks:
+
+- `PreToolUse` redirects Claude away from spawning expensive Explore agents.
+- `UserPromptSubmit` resets the per-turn token-savings counter.
+- `Stop` ingests new session transcript data and prints a cost receipt.
+- `SessionStart` reports index freshness (or a `tracedecay init` nudge) and, when the session restarts from compaction, injects the LCM context-recovery hint.
+- `PostToolUse` (matcher `Edit|MultiEdit|Write|NotebookEdit|Bash`) notifies the daemon so edits and shell commands trigger targeted incremental sync.
+
+The install also ships three read-only custom subagents into `~/.claude/agents/` — `code-explorer`, `code-health-auditor`, and `session-historian` — the same tracedecay subagents the Cursor plugin bundles. They are only replaced or removed when the file is tracedecay-managed; a same-named agent you authored yourself is left untouched. `tracedecay update-plugin` refreshes installed copies.
 
 ### Other agents