shraga 0.1.110 → 0.1.112

package/README.md

@@ -222,7 +222,8 @@ common ones:
 
 Beyond env vars, a typed config module in your data dir declares **global MCP servers**: the tools
 every user gets. (Per-user MCPs are added in the UI, and agent settings like model and engine live
-in `agent-config.json`, editable from the UI.) Shraga seeds it from a template on first run and
+in `agent-config.json`, editable from the UI; `"sdkResume": true` there opts the claude-code engine into
+SDK session resume — see `defaults/skills/platform.md`.) Shraga seeds it from a template on first run and
 gitignores it, so you just edit the seeded file:
 
 ```ts

package/defaults/skills/mcp-server.md

@@ -41,6 +41,8 @@ Add to `~/Library/Application Support/Claude/claude_desktop_config.json` (uses s
 }
 ```
 
+Optional bridge env: `MCP_BRIDGE_TIMEOUT_MS` (protocol calls, default 60s) and `MCP_BRIDGE_TOOL_TIMEOUT_MS` (`tools/call`, default 30min). Past the bound the bridge answers with a JSON-RPC error instead of hanging.
+
 ## Available MCP Tools
 
 | Tool | Description |

package/defaults/skills/platform.md

@@ -43,6 +43,7 @@ Users can prefix any message with `[directives]` to override settings. Parsed se
 | `effort:VALUE` | `low`, `medium`, `high`, `max` | Set reasoning effort level |
 | `model:VALUE` | Any alias | Override model |
 | `turns:VALUE` | Integer | Override max turns |
+| `resume:VALUE` | `on`, `off` | claude-code engine: resume the SDK session across turns instead of re-sending the history (prompt-cache savings). Overrides agent-config `sdkResume` for this conversation |
 
 ### Examples
 
@@ -68,6 +69,8 @@ Persistent settings that apply to all messages until changed:
 | Skill Discovery | On/Off | On |
 | System Prompt | Free text appended to system prompt | Empty |
 
+**SDK session resume (`sdkResume`, default off).** Not in the panel: set `"sdkResume": true` in `agent-config.json` (re-read every turn, no restart) or per conversation with `[resume:on]` / `PUT /api/sessions/:id/directives {"resume": true}`. When on, the claude-code engine sends the full context + history once (fresh query), then resumes that Claude Code session and sends only the new message, appending the current speaker's `user` section (every resume turn), changed context sections and messages other channels added AFTER the user text. It falls back to a fresh query — logged as `[claude] Cache: … path=fallback:<reason>` — on: `no-session`, `concurrent-run` (an earlier CLI process on the conversation, e.g. the run a steer took over, was still alive after 5s), `speaker-change` (a different person than the one whose turns the Claude Code session holds — keeps one user's private context out of another's turns; single-speaker conversations keep resuming), `engine-switch`, `account-change`, `reset`, `summary`, `history-diverged`, `drift`, `transcript-missing` (CLI cleanup removed it), `resume-failed` (the CLI couldn't load the session, e.g. "No conversation found" — retried fresh in the same turn; any other error, like a usage limit, surfaces as-is and keeps the mapping). Flag off logs `path=fresh`, a resumed turn `path=resume`.
+
 Directives override config and persist for the session. On a session's first turn, the resolved engine/model/turns/thinking are pinned to the session — reopening it from history resumes the exact same shape even if config defaults change later.
 
 ## Thinking / Reasoning

package/defaults/skills/self-aware.md

@@ -142,7 +142,7 @@ Your behavioral config (skills, MCPs, schedules, contacts, workspace, agent-conf
 - **Pull on demand**: `POST /api/data-sync/webhook` triggers a pull (GitHub webhook fires on every push)
 - **Change history**: `data/git-log.json` (gitignored, reconstructed) contains recent commit log — read it to understand what changed and when. Also available via `GET /api/data-sync/log`
 - **Recovery**: If data seems stale or missing, trigger a pull by restarting the service or calling the webhook
-- **Integrity audit**: `bun run src/server/integrity-audit.ts [git-ref]` compares current data/ against a baseline commit. Detects missing files, truncated content, degraded JSON (fewer entries in schedules/contacts/skills-defaults/api-keys), and invalid JSON. Runs automatically after every data-sync init — check logs for `[data-sync] ⚠️ DATA INTEGRITY`. Use manually when investigating suspected data loss.
+- **Integrity audit**: `bun run src/server/integrity-audit.ts [git-ref] [data-dir]` checks the files that changed between a baseline commit and HEAD (unchanged files are not re-examined — widen the ref to cover older commits). Detects missing files, truncated content, degraded JSON (fewer entries in schedules/contacts/skills-defaults/api-keys), and invalid JSON (.json over 16MB is skipped with a warning). Runs automatically after every data-sync init — check logs for `[data-sync] ⚠️ DATA INTEGRITY`. Use manually when investigating suspected data loss.
 
 ### Scheduler gating
 Schedules sync across all envs but only **execute** where `DATA_SYNC_SCHEDULER_ACTIVE=true` (prod). Inactive envs load and serve schedules with their real `enabled` flags — they just never fire them (no timers armed). Manual "run now" still works anywhere. Event triggers respect the same gate (`fireEvent` no-ops unless active) so a webhook can't double-fire across blue-green.