Hi,
We are seeing consistent 40-minute timeouts on our scaffold workflow starting today (Apr 21). Two of our runs hit the hard timeout before the agent could commit/push the scaffold branch. This was working reliably in ~15 minutes last week.
Affected runs (today):
09:26:47 ● Read module standard
09:29:11 ● Create feature branch (2m 24s gap)
09:31:50 ● List AVM module files (2m 39s gap)
09:34:00 ● Check existing renovate (2m 10s gap)
09:35:36 ● Create root outputs.tf
09:37:51 ● Create examples/data.tf (2m 15s gap)
09:45:21 ● Create storage_account.tftest.hcl (4m 36s gap)
09:47:42 ● Create README.md (2m 21s gap)
09:48:46 ● List all terraform files
09:59:14 ● Response was interrupted due to a server error. Retrying... (10m 28s gap)
10:06:01 ##[error] The action 'Execute GitHub Copilot CLI' has timed out after 40 minutes.
Working run for comparison (Apr 16, different module):
13:33:42 ● Read module standard
13:34:30 ● Check current git state (48s gap — longest)
13:34:36 ● Check AVM module variables (6s gap)
13:38:29 ● Create feature branch (LLM planning)
13:38:35 ● Create required directories (6s)
13:39:20 ● Create examples/provider.tf (22s)
... all gaps under 48s, completed in 11m 38s total
gh-aw version: v0.68.3
Direct timing comparison:
| Metric |
Apr 16 (working, 11m38s) |
Apr 21 (failing, 40m timeout) |
| LLM turn time (between tool calls) |
0–48 seconds max |
4–10 minutes |
| AVM module file reads |
6–19s (shell curl/gh api) |
37–71s each (get_file_contents MCP) |
| File creation tool calls |
0–22s |
1–4 minutes between each |
| Server errors |
None |
Response was interrupted due to a server error. Retrying... at 09:59 |
| Completed? |
✅ Yes |
❌ Both timed out |
Notable: last week the agent read AVM module files via shell (curl/gh api) taking 6–19s each. Today it routes through get_file_contents (MCP: github) taking 37–71s each — 10–12x slower. LLM response times are 20–30x worse.
Observed symptoms:
get_file_contents (MCP: github) calls taking 37–71 seconds each — reading individual files from Azure/terraform-azurerm-avm-res-storage-storageaccount via the GitHub MCP server. Example from run 1:
variables.tf → 71s
variables.storage.tf → 51s (then failed with path-not-found)
variables.storageaccount.tf → 62s
LLM turns taking 4–10 minutes — e.g. gap from 09:48:46 to 09:59:14 (10m28s) with a Response was interrupted due to a server error. Retrying... at the end.
Neither run completed — both exhausted the full 40-minute budget without reaching the git commit/push step.
Environment:
Lock file pins: ghcr.io/github/gh-aw-mcpg:v0.2.19, github-mcp-server:v0.32.0, gh-aw-firewall:0.25.20
gh-aw-actions@ba90f2186d7ad780ec640f364005fa24e797b360
No changes on our side between Apr 16 (working) and Apr 21 (failing)
Suspected cause: Copilot API rate limiting / degradation noted in #27339 (P1: rate limits) combined with elevated GitHub MCP server response times. The combination means a workflow that normally completes in 15 minutes cannot finish within the 40-minute cap.
Hi,
We are seeing consistent 40-minute timeouts on our scaffold workflow starting today (Apr 21). Two of our runs hit the hard timeout before the agent could commit/push the scaffold branch. This was working reliably in ~15 minutes last week.
Affected runs (today):
Working run for comparison (Apr 16, different module):
gh-aw version: v0.68.3
Direct timing comparison:
Notable: last week the agent read AVM module files via shell (curl/gh api) taking 6–19s each. Today it routes through get_file_contents (MCP: github) taking 37–71s each — 10–12x slower. LLM response times are 20–30x worse.
Observed symptoms:
get_file_contents (MCP: github) calls taking 37–71 seconds each — reading individual files from Azure/terraform-azurerm-avm-res-storage-storageaccount via the GitHub MCP server. Example from run 1:
variables.tf → 71s
variables.storage.tf → 51s (then failed with path-not-found)
variables.storageaccount.tf → 62s
LLM turns taking 4–10 minutes — e.g. gap from 09:48:46 to 09:59:14 (10m28s) with a Response was interrupted due to a server error. Retrying... at the end.
Neither run completed — both exhausted the full 40-minute budget without reaching the git commit/push step.
Environment:
Lock file pins: ghcr.io/github/gh-aw-mcpg:v0.2.19, github-mcp-server:v0.32.0, gh-aw-firewall:0.25.20
gh-aw-actions@ba90f2186d7ad780ec640f364005fa24e797b360
No changes on our side between Apr 16 (working) and Apr 21 (failing)
Suspected cause: Copilot API rate limiting / degradation noted in #27339 (P1: rate limits) combined with elevated GitHub MCP server response times. The combination means a workflow that normally completes in 15 minutes cannot finish within the 40-minute cap.