githubnext / githubnext/agentics
issue-arborist: `gh issue list` fails under DIFC proxy because /meta returns 403
- Langage dominant
- Makefile
- Étoiles
- 952
- Forks
- 139
- Merge moyen
- 6 j 10 h
- PR mergées (30 j)
- 2
Description
## Summary
The `Fetch issues data` step in `workflows/issue-arborist.md` uses `gh issue list`, which fails immediately on any GitHub Enterprise / DIFC-proxied runner because the `gh` CLI hits the `/meta` endpoint on startup and gets HTTP 403 from the proxy.
## Repro
Run the unmodified workflow in a repo whose runners are behind GitHub's corporate DIFC proxy (`localhost:18443` redirect for `GITHUB_API_URL` / `GITHUB_GRAPHQL_URL`, with a custom CA). The step fails on the first `gh` invocation:
```
HTTP 403: 403 Forbidden (https://localhost:18443/api/v3/meta)
##[error]Process completed with exit code 1.
```
Because this step runs before the MCP server starts, the rest of the job collapses with a misleading downstream symptom:
```
ERR_SYSTEM: rpc-messages.jsonl is present but zero bytes — MCP telemetry capture failed
(server may not have started or crashed before any RPC)
```
Example failed run from microsoft/testfx: https://github.com/microsoft/testfx/actions/runs/26294701799
## Root cause
`gh` CLI always probes `/meta` to verify the host before any subcommand. In DIFC-proxied environments the proxy blocks `/meta`, so every `gh` command fails — regardless of whether the actual GraphQL/REST endpoint would have worked.
## Suggested fix
Replace the `gh issue list` invocation in `workflows/issue-arborist.md` with a direct REST API call via `curl` to `GITHUB_API_URL/search/issues`, which doesn't trigger the `/meta` probe. We've been carrying this patch downstream since #8185 in microsoft/testfx and just had to re-apply it in microsoft/testfx#8507 after [re-syncing to upstream](https://github.com/microsoft/testfx/pull/8491).
The drop-in replacement (preserves the same JSON shape so the agent prompt is unchanged):
```yaml
- name: Fetch issues data
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
GH_AW_ORIGINAL_GITHUB_API_URL: ${{ github.api_url }}
GH_AW_GITHUB_REPOSITORY: ${{ github.repository }}
run: |
mkdir -p /tmp/gh-aw/issues-data
echo "⬇ Downloading the last 100 open issues (excluding sub-issues)..."
# Use REST API directly to avoid gh CLI /meta check blocked by DIFC proxy.
# State is normalized to uppercase (OPEN/CLOSED) to match gh CLI GraphQL output format.
curl -s \
-H "Authorization: Bearer ${GITHUB_TOKEN}" \
-H "Accept: application/vnd.github+json" \
--get \
--data-urlencode "q=repo:${GH_AW_GITHUB_REPOSITORY} is:issue is:open -is:sub-issue" \
--data-urlencode "sort=created" \
--data-urlencode "order=desc" \
--data-urlencode "per_page=100" \
"${GH_AW_ORIGINAL_GITHUB_API_URL}/search/issues" \
| jq '.items // [] | map({
number: .number,
title: .title,
author: {login: .user.login},
createdAt: .created_at,
state: (.state | ascii_upcase),
url: .html_url,
body: .body,
labels: [.labels[] | {name: .name}],
updatedAt: .updated_at,
closedAt: .closed_at,
milestone: (if .milestone != null then {title: .milestone.title} else null end),
assignees: [.assignees[] | {login: .login}]
})' \
> /tmp/gh-aw/issues-data/issues.json \
|| echo '[]' > /tmp/gh-aw/issues-data/issues.json
echo "✓ Issues data saved to /tmp/gh-aw/issues-data/issues.json"
echo "Total issues fetched: $(jq 'length' /tmp/gh-aw/issues-data/issues.json)"
```
## Scope
Any agentic workflow that shells out to `gh ` for prefetch (rather than going through the MCP server) will hit this same `/meta` 403. `issue-arborist` is the one biting us today, but it'd be worth auditing the others in the same way and either swapping them to `curl` or routing prefetches through the MCP server.
Happy to send a PR if useful.
Guide de contribution
Aucun guide de contribution indexé pour ce dépôt
Piste de recherche
Commencez dans workflows/issue-arborist.md à l’étape « Fetch issues data » et examinez comment sa sortie est consommée. Exécutez le workflow dans un environnement avec proxy DIFC, vérifiez que les données des issues sont écrites dans /tmp/gh-aw/issues-data/issues.json avec les champs attendus, et confirmez que le MCP server démarre sans l’erreur de télémétrie en aval. Vérifiez si d’autres workflows invoquent également gh via le shell pour effectuer une prélecture.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- github-actions, shell
- Domaine
- ci-cd, devops
- Type d'issue
- Bug
- Difficulté
- 3/5
- Temps estimé
- 1-2 jours
- Activité
- Calme
- Clarté
- Clairement spécifiée
- Accessibilité débutants
- 68/100