compact: compact worker をツール駆動マルチターンに再設計

段階 4〜9 を一括で実装：
- mark_read_required / add_reference / write_summary + read_file の 4 ツールで
  compact worker を駆動。結果は CompactWorkerContext に集約
- 新セッションの先頭を [summary, ...auto-read, references, ...retained] で構築
- デフォルトリファレンスは tracker.recent_files(5) から
- auto-read は compact_auto_read_budget で総量制限。超過は即エラー
- compact worker 自身は compact_worker_max_input_tokens で累計入力を制限
- 5 セクション要約フォーマットに system prompt を更新
- write_summary 未呼び出し / auto-read 空のときは 1 回追加プロンプトで促す

crates/manifest/src/config.rs

@@ -88,6 +88,10 @@ pub struct CompactionConfigPartial {
     #[serde(default)]
     pub compact_retained_tokens: Option<u64>,
     #[serde(default)]
+    pub compact_auto_read_budget: Option<u64>,
+    #[serde(default)]
+    pub compact_worker_max_input_tokens: Option<u64>,
+    #[serde(default)]
     pub provider: Option<ProviderConfigPartial>,
 }
 
@@ -244,6 +248,12 @@ impl CompactionConfigPartial {
             compact_retained_tokens: upper
                 .compact_retained_tokens
                 .or(self.compact_retained_tokens),
+            compact_auto_read_budget: upper
+                .compact_auto_read_budget
+                .or(self.compact_auto_read_budget),
+            compact_worker_max_input_tokens: upper
+                .compact_worker_max_input_tokens
+                .or(self.compact_worker_max_input_tokens),
             provider: merge_option(self.provider, upper.provider, ProviderConfigPartial::merge),
         }
     }
@@ -374,6 +384,12 @@ impl TryFrom<PodManifestConfig> for PodManifest {
                     compact_retained_tokens: c
                         .compact_retained_tokens
                         .unwrap_or(defaults::COMPACT_RETAINED_TOKENS),
+                    compact_auto_read_budget: c
+                        .compact_auto_read_budget
+                        .unwrap_or(defaults::COMPACT_AUTO_READ_BUDGET),
+                    compact_worker_max_input_tokens: c
+                        .compact_worker_max_input_tokens
+                        .unwrap_or(defaults::COMPACT_WORKER_MAX_INPUT_TOKENS),
                     provider: comp_provider,
                 })
             })

crates/manifest/src/defaults.rs

@@ -28,3 +28,20 @@ pub const COMPACT_RETAINED_TOKENS: u64 = 8000;
 /// is omitted. See the `PromptLoader` prefix addressing scheme for the
 /// `$insomnia/` / `$user/` / `$workspace/` namespaces.
 pub const DEFAULT_INSTRUCTION: &str = "$insomnia/default";
+
+/// Token budget for auto-read file contents injected into the new
+/// session after compaction. Limits how much raw file text the
+/// compact worker can pull into the compacted context via
+/// `mark_read_required`. See
+/// [`crate::CompactionConfig::compact_auto_read_budget`].
+pub const COMPACT_AUTO_READ_BUDGET: u64 = 8000;
+
+/// Cumulative input-token cap for the compact worker's own LLM
+/// calls. Exceeding this aborts the compact run (circuit-breaker
+/// path). See
+/// [`crate::CompactionConfig::compact_worker_max_input_tokens`].
+pub const COMPACT_WORKER_MAX_INPUT_TOKENS: u64 = 50_000;
+
+/// Number of recently-touched files fed to the compact worker as
+/// default references.
+pub const COMPACT_DEFAULT_REFERENCE_COUNT: usize = 5;

crates/manifest/src/lib.rs

@@ -201,6 +201,16 @@ pub struct CompactionConfig {
     #[serde(default = "default_compact_retained_tokens")]
     pub compact_retained_tokens: u64,
 
+    /// Aggregate token budget for auto-read file contents injected into
+    /// the compacted session by the compact worker.
+    #[serde(default = "default_compact_auto_read_budget")]
+    pub compact_auto_read_budget: u64,
+
+    /// Cumulative input-token cap for the compact worker's own LLM
+    /// calls. Exceeding this aborts the compact run.
+    #[serde(default = "default_compact_worker_max_input_tokens")]
+    pub compact_worker_max_input_tokens: u64,
+
     /// Optional provider for the compactor (summary) LLM.
     /// If omitted, the main provider is cloned via `clone_boxed()`.
     #[serde(default)]
@@ -216,6 +226,12 @@ fn default_prune_min_savings() -> u64 {
 fn default_compact_retained_tokens() -> u64 {
     defaults::COMPACT_RETAINED_TOKENS
 }
+fn default_compact_auto_read_budget() -> u64 {
+    defaults::COMPACT_AUTO_READ_BUDGET
+}
+fn default_compact_worker_max_input_tokens() -> u64 {
+    defaults::COMPACT_WORKER_MAX_INPUT_TOKENS
+}
 
 impl Default for CompactionConfig {
     fn default() -> Self {
@@ -225,6 +241,8 @@ impl Default for CompactionConfig {
             compact_threshold: None,
             compact_request_threshold: None,
             compact_retained_tokens: default_compact_retained_tokens(),
+            compact_auto_read_budget: default_compact_auto_read_budget(),
+            compact_worker_max_input_tokens: default_compact_worker_max_input_tokens(),
             provider: None,
         }
     }