Merge pull request #68 from protocol-security/post-codex-fixes

Post-Codex fixes: API 500 retry, dashboard label, setup.sh removal, weekly CI

Author: moodmosaic

.github/workflows/ci.yml

@@ -5,6 +5,8 @@ on:
     branches: [master]
   pull_request:
     branches: [master]
+  schedule:
+    - cron: '0 7 * * 1'
 
 jobs:
   shellcheck:
@@ -14,7 +16,7 @@ jobs:
       - name: ShellCheck
         run: |
           shellcheck -s bash --severity=warning \
-            launch.sh setup.sh dashboard.sh harvest.sh \
+            launch.sh dashboard.sh harvest.sh \
             costs.sh progress.sh \
             lib/*.sh lib/drivers/*.sh \
             tests/test_*.sh tests/test.sh

CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## 0.19.1 — 2026-04-13
+
+- **Retry on API 500 errors.** Claude Code driver now treats
+  `api_error`, `internal_error`, and HTTP 500 responses as
+  retriable, preventing agents from exiting on transient
+  Anthropic outages.
+- **Dashboard max-effort label.** `format_model` now displays
+  `(M)` for `effort: "max"` instead of `(m)`, disambiguating
+  it from medium.
+- **Remove setup.sh wizard.** Deleted the interactive `setup.sh`
+  and its tests. Copying a config from `tests/configs/` and
+  editing is simpler and better documented. Not treated as a
+  breaking change (would warrant 0.20.0) because no known
+  users rely on it.
+- **Weekly CI schedule.** CI now runs on a Monday morning cron
+  (`0 7 * * 1`) in addition to push/PR triggers.
+
 ## 0.19.0 — 2026-04-13
 
 - **Codex CLI driver.** New `codex-cli` driver

README.md

@@ -16,7 +16,6 @@ Based on the agent-team pattern from
 - tput (ncurses) — used by the dashboard
 - An Anthropic API key, OAuth token, or compatible endpoint
 
-Optional: `whiptail` for the interactive setup wizard TUI.
 For development: `shellcheck` for linting.
 
 ## Setup
@@ -62,10 +61,7 @@ see the changes on the next fetch.
 ## Quick start
 
 ```bash
-# Interactive setup (generates swarm.json).
-./setup.sh
-
-# Or create a swarmfile and launch.
+# Create a swarmfile and launch.
 SWARM_CONFIG=swarm.json ./launch.sh start --dashboard
 
 # Or place swarm.json in your repo root and launch.

USAGE.md

@@ -3,10 +3,7 @@
 ## Quick start
 
 ```bash
-# Interactive setup (generates swarm.json).
-./setup.sh
-
-# Or create a swarmfile manually and launch.
+# Create a swarmfile and launch.
 SWARM_CONFIG=swarm.json ./launch.sh start --dashboard
 ```
 
@@ -182,7 +179,6 @@ Unit tests (no Docker or API key):
 ./tests/test_harness.sh        # Stat extraction.
 ./tests/test_harvest.sh        # Harvest git ops.
 ./tests/test_launch.sh         # Launch logic.
-./tests/test_setup.sh          # Setup wizard.
 ```
 
 ## Post-processing

VERSION

@@ -1 +1 @@
-0.19.0
+0.19.1

dashboard.sh

@@ -210,7 +210,9 @@ format_tps() {
 format_model() {
     local m="${1:-unknown}" effort="${2:-}"
     if [ -n "$effort" ]; then
-        printf '%s (%s)' "$m" "${effort:0:1}"
+        local e="${effort:0:1}"
+        [ "$effort" = "max" ] && e="M"
+        printf '%s (%s)' "$m" "$e"
     else
         printf '%s' "$m"
     fi

lib/drivers/claude-code.sh

@@ -123,8 +123,10 @@ agent_is_retriable() {
         "$logfile" 2>/dev/null && echo "rate_limited" && return
     grep -q '"Too many requests"\|"rate limit"' \
         "$logfile" 2>/dev/null && echo "rate_limited" && return
+    grep -q '"api_error"\|"internal_error"\|"Internal server error"\|"api_error.*500"\|"500.*api_error"' \
+        "$logfile" 2>/dev/null && echo "server_error" && return
     if [ -f "${logfile}.err" ]; then
-        grep -qi 'rate.limit\|too many requests\|overloaded' \
+        grep -qi 'rate.limit\|too many requests\|overloaded\|internal.server.error\|api.error.*500\|500.*error' \
             "${logfile}.err" 2>/dev/null && echo "rate_limited" && return
     fi
     return 0