fix(chat/runs): abort headless run when platform-api skill is missing

[sweetmantech]

What

After install + discovery in provisionRunSession, fail closed: if recoup-platform-api-access isn't among the discovered skills, throw. The caller (handleStartChatRun) already maps a provisionRunSession throw to a 5xx and revokes any minted ephemeral key — so the run aborts cleanly instead of proceeding.

Why

Installing skills in the headless path (the stacked PR below) is necessary but not sufficient: a best-effort install can still silently fail (network/registry/timeout), leaving the agent with no skill tool. Today that degrades to the agent guessing API endpoints and fabricating a report it then emails to a customer (recoupable/chat#1822).

This makes the failure observable and safe: a missed run is recoverable; a fabricated report sent to a label/manager is not.

Behavior choice: this is fail-closed (abort). The alternative — log+metric and send a degraded "no data" email — is also defensible. Aborting was chosen because a fabricated send is the worst outcome. Happy to switch to warn-only or add a retry if preferred.

Tests (TDD, RED→GREEN)

Extends lib/chat/runs/__tests__/provisionRunSession.test.ts:

• aborts when the platform API-access skill is missing after discovery — discoverSkills → [] ⇒ provisionRunSession rejects with /recoup-platform-api-access/ (was RED; now GREEN).
• Existing success/best-effort tests updated so discovery returns the skill.

lib/chat/runs suite green (20 passed), eslint clean, zero new tsc errors in changed files.

Stacking

Stacked on #722 (fix/headless-install-global-skills) — base is that branch so this PR's diff is only the abort check. Retarget to test after #722 merges.

Refs recoupable/chat#1822 · merge order: skills#65 → api#722 → this · base fix/headless-install-global-skills

🤖 Generated with Claude Code

---

Summary by cubic

Abort headless chat runs if the required recoup-platform-api-access skill isn’t present after install + discovery. This prevents ungrounded runs and avoids sending fabricated reports.

• Bug Fixes
  • Fail-closed guard in provisionRunSession: throw when recoup-platform-api-access isn’t discovered; caller maps to 5xx and revokes the ephemeral key.
  • Tests: default discovery includes the skill; added missing-skill rejection case.

^{Written for commit b3b40ee. Summary will update on new commits.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
api	Ready	Preview	Jun 30, 2026 1:54am

[coderabbitai]

Warning

Review limit reached

@sweetmantech, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 43 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 178ddfd5-2431-4a94-b1d4-bd0e31e766f6

📥 Commits

Reviewing files that changed from the base of the PR and between 8a3d084 and b3b40ee.

⛔ Files ignored due to path filters (1)

• lib/chat/runs/__tests__/provisionRunSession.test.ts is excluded by !**/*.test.*, !**/__tests__/** and included by lib/**

📒 Files selected for processing (1)

• lib/chat/runs/provisionRunSession.ts

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/headless-reliability-skill-check

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[cubic-dev-ai]

1 issue found across 2 files

Confidence score: 3/5

• In lib/chat/runs/provisionRunSession.ts, the abort/error path can leave a sandbox/session persisted as ACTIVE when required skill discovery fails, which risks orphaned resources and inconsistent run state after merge; add explicit cleanup/failed-session handling before throwing, or perform skill validation before writing ACTIVE state.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="lib/chat/runs/provisionRunSession.ts">

<violation number="1" location="lib/chat/runs/provisionRunSession.ts:114">
P2: Abort path leaks a provisioned active sandbox/session when required skill discovery fails. Add cleanup/failed-session handling before throwing, or move the validation before persisting ACTIVE state if possible.</violation>
</file>

Architecture diagram

sequenceDiagram
    participant Client as Chat Client
    participant Handler as handleStartChatRun
    participant Provision as provisionRunSession
    participant Installer as installSessionGlobalSkills
    participant Discoverer as discoverSkills
    participant SkillList as Discovered Skills
    participant SessionDB as sessions table

    Note over Client,SessionDB: HEADLESS RUN PROVISION FLOW

    Client->>Handler: Start headless chat run
    Handler->>Provision: provisionRunSession(accountId, title)

    Provision->>Provision: Create session record
    
    Provision->>Installer: Install global skills (best-effort)
    
    alt Install fails (network/registry/timeout)
        Installer-->>Provision: Error thrown
        Note over Provision: Continues anyway (best-effort)
    end

    Provision->>Discoverer: discoverSkills(sandbox)
    Discoverer-->>Provision: List of discovered skills

    Note over Provision: REQUIRED_PLATFORM_API_SKILL = "recoup-platform-api-access"

    alt Skill "recoup-platform-api-access" IS in discovered list
        Provision-->>Handler: Return ProvisionedRunSession
        Handler->>Handler: Proceed with run
        Handler->>Client: 200 OK + run results
    else Skill "recoup-platform-api-access" NOT in discovered list
        Note over Provision: Fail closed - abort immediately
        Provision-->>Handler: Throw Error
        Handler->>Handler: Map to 5xx error
        Handler->>Handler: Revoke ephemeral key
        Handler-->>Client: 500 Internal Server Error
        Note over Client: Run aborted - recoverable state
    end

    Note over Handler,Client: DECISION: Fail-closed chosen over<br/>degraded "no data" email because<br/>fabricated report is worst outcome

Loading

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: Abort path leaks a provisioned active sandbox/session when required skill discovery fails. Add cleanup/failed-session handling before throwing, or move the validation before persisting ACTIVE state if possible.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At lib/chat/runs/provisionRunSession.ts, line 114:

<comment>Abort path leaks a provisioned active sandbox/session when required skill discovery fails. Add cleanup/failed-session handling before throwing, or move the validation before persisting ACTIVE state if possible.</comment>

<file context>
@@ -101,6 +105,17 @@ export async function provisionRunSession({
+  // A missed run is recoverable; a fabricated report sent to a customer is not.
+  // The caller maps this throw to a 5xx and revokes any minted ephemeral key.
+  if (!skills.some(skill => skill.name === REQUIRED_PLATFORM_API_SKILL)) {
+    throw new Error(
+      `[provisionRunSession] required skill '${REQUIRED_PLATFORM_API_SKILL}' unavailable after install/discovery — aborting to avoid an ungrounded run`,
+    );
</file context>

[sweetmantech]

Closing per YAGNI. recoupable/skills#65 + #722 are the actual fix (the skill points at a live host + headless runs now install it, so the agent gets the skill tool). This PR's downstream fail-closed abort guards a should-not-happen path (best-effort install silently failing), and a hard abort would halt all scheduled emails on a transient skills-registry hiccup — a new risk added speculatively.

If we ever observe skill-less runs in practice, the lighter follow-up is observability (log/metric on a zero-skill provision), not a hard abort. Decision recorded on recoupable/chat#1822.