feat: improve generated agent prompts

TASKS.md

@@ -116,7 +116,7 @@ Acceptance Criteria:
 
 ## TASK-007 — Improve generated agent prompts
 
-Status: Todo
+Status: Done
 
 Goal:
 Make `rdb prompt` produce smaller, more direct prompts for Claude Code/local LLM agents.
@@ -130,21 +130,59 @@ Acceptance Criteria:
 - Prompt limits scope to one small implementation step
 - Add/update tests
 
-## TASK-012 — Agent telemetry
+## TASK-012 — Agent telemetry foundation
 
 Status: Todo
 
 Goal:
-Capture actual agent actions rather than inferring behaviour
-from documentation and run logs.
+Create a minimal telemetry writer for future agent activity tracking.
 
-Examples:
+Implementation Gap:
+There is currently no persistent session log for recording agent activity. Guardrails infer behaviour from markdown and RUN_LOG.md instead of structured telemetry.
 
-- file reads
-- file writes
-- command execution
-- test execution
-- git activity
+Acceptance Criteria:
 
-Output:
-.rdb/session-log.jsonl
+- Add a telemetry module
+- Create `.rdb/session-log.jsonl` when recording an event
+- Record JSONL events with:
+  - timestamp
+  - event_type
+  - target
+  - details
+- Add tests for writing and appending telemetry events
+- Do not integrate telemetry into every CLI command yet
+
+## TASK-013 — Log CLI command execution
+
+Status: Todo
+
+Goal:
+Record rdb CLI command execution into telemetry.
+
+Implementation Gap:
+Telemetry writer exists, but CLI commands do not yet record command execution events.
+
+Acceptance Criteria:
+
+- Record command execution events
+- Include command name
+- Include success/failure where practical
+- Add/update tests
+
+## TASK-014 — Use telemetry in guardrails
+
+Status: Todo
+
+Goal:
+Make `rdb guardrails` use structured telemetry when available.
+
+Implementation Gap:
+Guardrails currently rely on markdown and file mtimes rather than `.rdb/session-log.jsonl`.
+
+Acceptance Criteria:
+
+- Read telemetry events when present
+- Detect repeated reads from telemetry
+- Detect repeated commands from telemetry
+- Fall back gracefully if telemetry is missing
+- Add/update tests