Skip to content

feat/agentcore 3 aws - #6773

Closed
kyleseaman wants to merge 2 commits into
feat/agentcore-2-principalfrom
feat/agentcore-3-aws
Closed

kyleseaman wants to merge 2 commits into
feat/agentcore-2-principalfrom
feat/agentcore-3-aws

Conversation

@kyleseaman

Copy link
Copy Markdown
Contributor

Problem / Motivation

Why it matters

What changed (motivation → approach → change)

Tests

Manual verification

Screenshots / video

Related Issues

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Land the inert AWS extra and launch-path identity so later Gateway
and login PRs have something to attach to. Default-off: no policy or
env posture means no boto3, no extra install, no adapter swap.

Workload gateway_mcp_spec returns the https hostname and does not
import the SigV4 proxy; that rewrite belongs in the next PR.
resolved_* reads the composed ceiling first. Workload Allow is
GetWorkloadAccessToken only; UserId stays on the login Deny.
Inspect is Get/List/GetTarget — no Synchronize. Dashboard launch
stays none.
@kyleseaman
kyleseaman requested review from a team and pepmach and removed request for a team August 29, 2026 11:40
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Aug 29, 2026
@kyleseaman

Copy link
Copy Markdown
Contributor Author

Opened by accident while trying to rename the AgentCore stack off the cursor/ prefix. The live stack remains #6594–#6602. Closing this duplicate.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Aug 29, 2026
@kyleseaman kyleseaman closed this Aug 29, 2026
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Aug 29, 2026
@github-actions

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — 🟡 CONCERNS

Design-level review of 9db6b9d0882463c67570da31ce9c6bacfaffd25d — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

Sound, well-staged opt-in adapter and boundary design — but a 2,600-line IAM-sensitive diff shipped with a completely empty PR description.

Watch

  • Empty description on a security-sensitive change. The PR body is the untouched template — no problem, approach, tests, or manual-verification statement. Description↔diff fidelity cannot be audited for a diff that rewrites the launcher IAM document, adds a second permissions boundary, and swaps a platform adapter at boot; the merge record justifies none of it. Fill it in before merge.
  • The launcher policy widens unconditionally. policy_document() now always carries AgentCoreWorkloadIdentityControlPlane (Create/Delete/Update WorkloadIdentity on kirocrew*), so every operator who re-applies the "least-privilege" policy grants their launch credential AgentCore control-plane verbs they may never use — a leaked credential can then delete a crew's workload identity even in a fleet that never opted in.
  • Roughly a third of agentcore_aws.py has no caller in this PR (ensure_extra, apply_agentcore_runtime, probe_workload_identity, extra_snapshot) — it targets the "later stack PR" Settings surface, and cloud.md already describes the force-install behavior in the present tense. If the next PR's shape shifts, this pre-built surface and the spec drift silently.

Suggestions

  • Gate the AgentCore statements in the launcher document behind an opt-in (iam-policy --agentcore), mirroring the iam-boundary --agentcore split, so the default document stays byte-stable and free of bedrock-agentcore verbs.

[DESIGN-REVIEWED] 9db6b9d

@github-actions

Copy link
Copy Markdown
Contributor

Opus 4.8 Review — ✅ no blocking findings

Reviewed 9db6b9d0882463c67570da31ce9c6bacfaffd25d — this comment is updated in place on each push.

Review details

Both candidates die under falsification:

  • Candidate 1 (resolved_workload_name env precedence): honoring an explicitly-set KIROCREW_AGENTCORE_WORKLOAD_NAME over an unnamed home posture is a debatable precedence choice, not an observable wrong outcome — no crash, no security boundary crossed. The precondition set (no loaded ceiling + home-authored posture with empty name + stale env name) is not shown to co-occur in code I opened. Fails (c) and arguably (a). Below bar.

  • Candidate 2 (Gateway-URL validator vs CFN AllowedPattern): the client validator lives in agentcore_schema.py, which this PR does not change; the divergence merely means a query-string URL passes the pre-check and is then refused by CloudFormation's AllowedPattern — fails closed with an error, no shell reaches an unvalidated value (the on-box systemd env is populated from the CFN-validated parameter). Pure UX/consistency divergence, not a security hole, crash, or data loss. Below bar.

No new grounded finding surfaced during falsification: the launcher policy correctly withholds InvokeGateway/GetWorkloadAccessToken*, the successor boundary is reference-only (no CreatePolicy mint), the content-fixed boundary verification fails closed, and the systemd %-escaping plus CFN charset block shell/credential injection into the unit file.

No findings.

[OPUS-REVIEWED] 9db6b9d

Verdict parsed from the review's SHA-scoped output markers for commit 9db6b9d0882463c67570da31ce9c6bacfaffd25d.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 9db6b9d0882463c67570da31ce9c6bacfaffd25d: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — 🟡 CONCERNS

Premise-level review of 9db6b9d0882463c67570da31ce9c6bacfaffd25d — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All evidence gathered; counts verified. Final review follows.

First-Principles-Verdict: CONCERNS

A quarter of this surface — self-pip-install, probes, hot-apply, dashboard launch plumbing — has zero consumers; two doc sentences describe code that doesn't exist.

What this change ships

Intent: let an operator give a cloud-launched crew an Amazon Bedrock AgentCore workload identity (CFN resource + posture-scoped IAM) with a default-off AWS identity adapter — an ADDITION. The PR description is an untouched template, so every item below is undeclared.

  1. cloud launch --agentcore-posture creates the identity + instance grant at deploy — justified
  2. --agentcore-gateway-url writes an existing Gateway URL onto the box — justified
  3. iam-policy --instance --posture prints the instance fragment — CLI justified; HTTP twin zero consumers
  4. iam-boundary --agentcore pre-creates the successor boundary — justified (immutable-ceiling invariant)
  5. New kirocrew[agentcore] extra + install.sh --agentcore — justified
  6. Boot swaps the AWS adapter when ceiling/env opts in, default off — justified (spec updated same commit)
  7. Launcher policy grows control-plane verbs; CreateRole accepts either boundary — justified
  8. Dashboard launch API accepts/persists agentcore fields while pinning posture "none" — zero consumers
  9. ensure_extra/extra_snapshot/probe_workload_identity/apply_agentcore_runtime/probe_instance_invoke_gateway — zero consumers
  10. Black reformat of five cloud modules + three test files — rides along

Watch

  • Empty description on a change adding IAM verbs, an install flag, and a boot-time adapter swap: nothing is declared, the checklist is unticked.
  • docs/system-specs/modules/cloud.md states present-tense behavior the diff doesn't ship: "The Settings → Remote Crew setup tab copies them as two separately labeled buttons" (no website/ change; grep instance_policy|instance=1 in website/src → 0; client.ts:2230 fetches only {policy}), and "also force-installs that extra into the running gateway interpreter" (ensure_extra callers in src/ → 0).
  • GET /api/cloud/iam-policy?instance=1 duplicates the CLI iam-policy --instance for a UI the docs themselves defer to "a later stack PR" — 0 consumers.
  • test/test_spawn_audit.py pre-declares agentcore_aws.py::ensure_extra benign — a security exemption for a spawn no code path reaches.

Subtractions

  • Delete the dashboard/LaunchJob agentcore plumbing — LaunchJob.agentcore_posture/agentcore_gateway_url, the provision() kwargs, the persisted JSON keys, and body parsing in handlers_cloud.py. The store's sole production caller (handlers_cloud.py, grep cloud_launch_store.create → 1) hard-pins posture "none", the website sends neither field (grep → 0), and a URL without posture is inert on the box. Keep only the 400 on an attempted posture key.
  • Delete probe_instance_invoke_gateway (src/kiro_crew/cloud/iam.py:337) — 0 consumers; the "login-posture withhold" its docstring cites does not exist in src/.
  • Defer ensure_extra, extra_snapshot, probe_workload_identity, apply_agentcore_runtime (plus the spawn-audit row) to the Settings-PUT PR that calls them — 0 consumers each (grep across src/ → definitions only); and drop the no-effect posture parameter on agentcore_boundary_policy_document (its own docstring: "the document is the same union for both") with zero-consumer agentcore_boundary_policy_json.

[FIRST-PRINCIPLES-REVIEWED] 9db6b9d

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 9db6b9d0882463c67570da31ce9c6bacfaffd25d and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 9db6b9d

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 9db6b9d0882463c67570da31ce9c6bacfaffd25d: <one-sentence reason>

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant