feat(agent-loop): add resume command

- Add agent-loop:resume command to resume loop from persisted state file.
- Register agent-loop:resume command in Codex manifest.
- Update documentation in README.md and CLAUDE.md.

Co-Authored-By: duyetbot <duyetbot@users.noreply.github.com>

Author: duyet

agent-loop/.codex-plugin/plugin.json

@@ -17,7 +17,8 @@
       "agent-loop:stop",
       "agent-loop:status",
       "agent-loop:triage",
-      "agent-loop:autoreview"
+      "agent-loop:autoreview",
+      "agent-loop:resume"
     ]
   },
   "interface": {

agent-loop/CLAUDE.md

@@ -34,6 +34,7 @@ agent-loop/
 | `/agent-loop:status`     | Check loop status and history         |
 | `/agent-loop:triage`     | One-shot repo queue triage            |
 | `/agent-loop:autoreview` | One-shot PR review and fix            |
+| `/agent-loop:resume`     | Resume the loop from persisted state  |
 
 ### Skills
 

agent-loop/README.md

@@ -41,6 +41,7 @@ That makes it easy to parallelize + steer work as needed.
 | `/agent-loop:status`     | Check loop status and history         |
 | `/agent-loop:triage`     | One-shot repo queue triage            |
 | `/agent-loop:autoreview` | One-shot PR review and fix            |
+| `/agent-loop:resume`     | Resume the loop from persisted state  |
 
 ## Skill Sources
 

agent-loop/commands/resume.md

@@ -0,0 +1,37 @@
+---
+allowed-tools: Bash, Read, Write, Task, Agent
+description: Resume the agent-loop from the last persisted state in .agent-loop/state.json. Wakes every 5 minutes, triages repos, dispatches work to parallel agent threads, monitors progress, and reports results.
+---
+
+# Agent Loop Resume
+
+Resume the continuous agent-loop from the last saved state in `.agent-loop/state.json`.
+
+The loop will load the state file to recover statistics (cycle count, items processed, PRs merged) and continue the cycle:
+1. **Load State** — parse `.agent-loop/state.json` to restore loop context
+2. **Wait** — sleep for configured interval (default: 5 min)
+3. **Triage** — scan repo queues for actionable items
+4. **Dispatch** — route autonomous work to parallel agent threads
+5. **Monitor** — track thread progress and outcomes
+6. **Report** — log cycle summary
+
+## Usage
+
+```bash
+/agent-loop:resume
+/agent-loop:resume --interval 300
+```
+
+## Options
+
+- `--interval <seconds>`: Sleep between cycles (default: 300)
+- `--scope <current|all|org>`: Triage scope (default: current)
+- `--max-concurrency <n>`: Max parallel threads (default: 3)
+- `--repos <owner/repo,...>`: Comma-separated repos for broad scope
+- `--dry-run`: Preview what the loop would do without executing
+
+## State File
+
+State is recovered from and persisted to `.agent-loop/state.json` in the repo root.
+Use `/agent-loop:status` to inspect current state.
+Use `/agent-loop:stop` to gracefully shut down.