Scaffold fullsend-github-ro profile lacks the GraphQL endpoint the fleet profile has; generated agents fail on `gh --json`

[waynesun09]

What happened

While reviewing the fullsend agent new reference example (fullsend-ai/agents#1167, part of #6966), a generated agent whose first step is gh pr view --json ... could not run: the profile the generator copies from fullsend's embedded scaffold, internal/scaffold/fullsend-repo/profiles/fullsend-github-ro.yaml, has no GraphQL endpoint. The fleet's copy in fullsend-ai/agents profiles/fullsend-github-ro.yaml does, and its own comment says "Endpoints mirror fullsend-github-code.yaml — keep both in sync".

Consequence: any custom agent that gets its profiles from the scaffold (every agent new user, and the CI workspace-preparation layer for per-repo installs) is blocked at egress on gh ... --json, gh pr view, gh issue view and anything else that goes through api.github.com/graphql, with a policy denial that looks like an agent bug.

What should happen

The scaffold profile should carry the same endpoint set as the fleet profile, and the two should have one source of truth. This is the same drift as #6981 (Vertex profile: credentials flag) and #6971 (Vertex profile: pi / claude.exe binaries), on a different profile with a larger blast radius.

Workaround used in the example: REST calls only (gh api repos/{owner}/{repo}/pulls/{n}/files), no --json.

Found by the multi-agent review of fullsend-ai/agents#1167. Related: #6966, #6971, #6981.

[fullsend-ai-triage]

🤖 Finished Triage · ✅ Success · Started 3:51 PM UTC · Completed 3:56 PM UTC

Commit: fecb2f0 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $0.80

[fullsend-ai-triage]

Severity: high — every agent generated via fullsend agent new is affected; the workaround (REST-only calls) limits what generated agents can do.

Recommended fix

Add the missing GraphQL endpoint block to internal/scaffold/fullsend-repo/profiles/fullsend-github-ro.yaml, matching the fleet copy:

  - host: api.github.com
    port: 443
    protocol: graphql
    access: read-only
    enforcement: enforce
    path: "/graphql"

Proposed test

// Assert the scaffold fullsend-github-ro profile contains
// a GraphQL endpoint for api.github.com.
func TestGitHubROProfileHasGraphQLEndpoint(t *testing.T) {
    profile := loadProfile(t, "fullsend-github-ro.yaml")
    found := false
    for _, ep := range profile.Endpoints {
        if ep.Host == "api.github.com" && ep.Protocol == "graphql" {
            found = true
            assert.Equal(t, "/graphql", ep.Path)
        }
    }
    assert.True(t, found,
        "scaffold fullsend-github-ro profile must include GraphQL endpoint")
}

Related issues

• scaffold: lint drift between internal/scaffold/fullsend-repo/profiles and the fullsend-ai/agents profiles #6978 — proposes a lint check to catch all scaffold-fleet drift (prevention)
• Provider profiles (.fullsend/profiles/*.yaml) have no update path — allowlist fixes never reach installed repos #6990 — profiles have no update path once installed (systemic)
• sandbox: scaffold Vertex profile lacks **/claude.exe, so every 0.40.0-image Claude run fails with policy_denied #6971, Scaffold fullsend-vertex-ai profile allows uninspected credentials on api.anthropic.com; the fleet profile does not #6981 — prior instances of the same class of drift on other profiles (both closed)

---

Labels: Sandbox profile defect blocking generated agents, high impact on agent-new users.

---

Next steps:

• /fs-code — agent creates a PR to implement this issue
• /fs-code <your instruction> — agent implements with your specific guidance

[fullsend-ai-coder]

🤖 Finished Code · ✅ Success · Started 3:58 PM UTC · Completed 4:06 PM UTC

Commit: fecb2f0 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $1.94