[realAlexBarge]

The skill has shipped in the plugin since 2026-09-05 but was missing from the skills table. This adds its row. Most of the diff is Prettier widening the first column.

[solution-architecture-reviewer]

solution-architecture-reviewer bot: 3 item(s) for a human

Not approving, not blocking. Something below is worth a look, but none of it is the solution-architecture-reviewer bot's call to settle: a human decides.

Reviewed 1 changed file(s).

Axis assessment

• 🟡 Manifest drift — The PR description claims the skill self-hosted-llm-batch-inference has shipped in the plugin since 2026-09-05 but was simply missing from the table; this diff only shows the README.md table, so there is no visibility into the actual skills/ directory to confirm the skill really already exists and matches this description. If the skill directory does not already exist, this would be manifest drift (documenting a skill not yet shipped), but nothing in the provided diff contradicts the claim.
• 🟡 Upstream duplication (advisory) — The new README row describes self-hosted-llm-batch-inference as covering 'offline inference for a 1B to 8B open-weight model on AWS Batch with Amazon ECS Managed Instances Spot: no idle GPU between runs, and automatic recovery from Spot interruption.' This is plausibly startup-shaped (cost/idle-GPU framing) but the skill's own SKILL.md text was not supplied, so I cannot verify its actual description against aws-compute's Spot/ASG coverage or aws-containers's ECS Managed Instances coverage; this is a judgment criterion 2 defers to human review and I lack the skill body to make a concrete claim.
• 🟡 Scope (startup-specific) (advisory) — The table description frames the skill as a specific technical problem (batch GPU inference cost/idle avoidance) which looks startup-shaped per criterion 1, but since only the README row is in the diff and not the skill's SKILL.md, a reviewer should still confirm the full skill body doesn't drift into general-purpose Batch/ECS guidance that would read identically for an enterprise.
• ✅ Contradiction — No conflicting guidance introduced; this is a table-row addition and column widening in README.md.
• ✅ Staleness — The new row references self-hosted-llm-batch-inference with AWS Batch and Amazon ECS Managed Instances Spot, both current services; no sunset-service mention appears in the diff.
• ✅ Activation quality (advisory) — The added description for self-hosted-llm-batch-inference is specific (model size range, AWS Batch, ECS Managed Instances Spot, idle-GPU and Spot-interruption framing) and does not overlap textually with sibling rows agentcore-patterns, cost-regression-check, or hcls-stack-check.
• ✅ Injection posture (advisory) — No new agent or untrusted-input-consuming skill is introduced in this diff; it is a documentation table edit.
• ✅ PR hygiene (advisory) — The PR description discloses that most of the diff is Prettier widening the first column and that the change simply adds a missing row for an already-shipped skill; this matches solution-architecture/CONTRIBUTING.md's carve-out for corrections to existing content ('wording that misstates what the skill does') not requiring an RFC, which is a disclosed and plausible rationale.
• ✅ Authoring conventions (advisory) — The table-formatting change in README.md matches the described Prettier/dprint widening behavior in the PR description and introduces no departure from documented conventions in solution-architecture/CONTRIBUTING.md.

🔴 concern · 🟡 borderline · ✅ clear. Advisory axes are reported but never block, since the contribution gate reserves those criteria for human reviewers.

_{Reviewed 1 file(s). Model: us.anthropic.claude-sonnet-5. Findings below 0.6 confidence are surfaced as borderline rather than hedged. Judgment-only: formatting, schema, and security scanning are covered by existing CI.}