---

name: adversarial-beam description: Use this skill to gate Elixir/OTP systems on BEAM runtime architecture with a pass/fail adversarial review — a single blind reviewer subagent judges a diff or file set against 23 decidable rules covering supervision and failure design (restart strategy vs child dependencies, blocking init, durability in terminate, restart amnesia, unsupervised fire-and-forget), backpressure (cast on externally-driven ingest, push pipelines without demand, silenced or retried call timeouts, unbounded fan-out), event delivery semantics (PubSub treated as durable, jobs enqueued outside the creating transaction, non-idempotent consumers under at-least-once delivery, cross-source ordering, telemetry handlers that block or detach), shared-state races (check-then-act on ETS/Registry, hot reads through one mailbox, persistent_term churn), distribution (global singletons in netsplits, node-local uniqueness assumed cluster-wide), and runtime mechanics (wall-clock durations, sub-binary leaks, minted atoms). Verdicts only, never fixes.

Adversarial BEAM Gate

A BEAM runtime-architecture review gate for Elixir systems — pass/fail: a single blind reviewer subagent judges the work against this gate's rules with an adversarial mandate, and the work passes only when every rule is PASS or N/A. This skill renders verdicts; it never fixes the work.

The rules target one failure mode: code that assumes guarantees the runtime never made — that restarts restore data, that mailboxes push back, that broadcasts arrive, that delivery happens exactly once and in order, that a lookup is still true at the following insert, that node-local uniqueness holds cluster-wide, that the wall clock only moves forward. These bugs compile cleanly, pass unit tests, and surface as production incidents. Each rule carries an Evidence of violation paragraph so a reviewer can decide PASS/FAIL/N/A from artifact evidence alone. The sibling gate adversarial-elixir judges paradigm fit (OO habits ported onto the BEAM); this gate judges runtime semantics — run both for full coverage.

When to Apply

• An Elixir feature that supervises processes, consumes events/jobs/streams, broadcasts over PubSub, or shares state via ETS/Registry is about to merge and needs an objective PASS/FAIL on its runtime assumptions.
• An agent (Claude, Codex) authored OTP-shaped code and you want an independent check that it did not assume durable broadcasts, exactly-once delivery, ordered cross-process messages, or restart-survivable in-memory state.
• A system is being prepared for clustering or higher load, and the single-node/low-traffic assumptions baked into it need to be surfaced as verdicts.
• An incident postmortem fixed one instance of a delivery/overload bug and you want the same class hunted across the affected area.

Do not apply to non-Elixir codebases (the reviewer prompt's precondition aborts with "GATE NOT APPLICABLE"), or when the user wants explanations and refactors rather than a verdict. The dist-* category goes N/A when the application demonstrably never clusters; Oban/Broadway appear in rules as canonical remedies but any mechanism with the same property passes — the reviewer prompt carries the applicability axes.

Review Protocol

Follow these steps exactly — the gate's value is that every review runs the same way.

1. Identify the target. Pin down exactly what is under review (a diff, a set of files, a PR) and note the ref/paths so the review runs against an unambiguous, fixed target. Record the stack facts (from mix.exs and config/) — Elixir/OTP requirement, clustering present or absent, Ecto/Oban/Broadway/PubSub presence, telemetry attachments. Include the repo root in the target description — several rules must search beyond the diff for sibling init bodies, rehydration paths, durable mechanisms, and clustering config (the reviewer prompt lists them).
2. Load the rules. Read references/_sections.md and every rule file in references/ (all sup-*.md, load-*.md, evt-*.md, state-*.md, dist-*.md, mech-*.md files).
3. Compose the reviewer prompt. Fill references/reviewer-prompt.md with the rules, the target, and the stack facts. The composed prompt must be fully self-contained — a reviewer sees no conversation history, so nothing may refer to context outside the prompt.
4. Dispatch one blind reviewer. Launch a single Task subagent whose entire input is the composed prompt — no conversation context, no commentary alongside it.
5. Render fail-closed. The reviewer's structured output is the verdict — there is no merge step. Overall verdict is PASS only when every rule is PASS or N/A; any single FAIL fails the gate. Never average, weigh severity, or waive a rule — a "minor" FAIL is a FAIL. If the reviewer returns "GATE NOT APPLICABLE" (not an Elixir target), stop and report that instead of a verdict.
6. Render the verdict. Fill assets/templates/verdict.md. On FAIL, aggregate the reviewer's "missing for PASS" suggestions into the fix list, each with its location, ordered by category importance. Every rule whose final result is FAIL must appear in the fix list with a change concrete enough to apply as written — if the reviewer's suggestion only restates the violation, derive the fix from the rule's Correct example before rendering.

If the same rule flips verdicts across re-reviews of an unchanged target, or a human reads the evidence and overrides the verdict, that is a decidability bug in the rule — record it in gotchas.md and sharpen the rule; do not override the gate.

Verdict Format

The reviewer returns, per rule: PASS | FAIL | N/A, evidence (file:line or a quote — required for PASS as well as FAIL), and for every FAIL, the fix that flips the rule to PASS once applied — the named change plus its location, never a restatement of the violation. The final report follows assets/templates/verdict.md.

Rule Categories

#	Category	Prefix	The wrong assumption it gates
1	Supervision & Failure Design	sup-	"The supervisor restarts it, so we recover" — strategies ignoring child dependencies, blocking init, durability in terminate/2, restart amnesia, unawaited/unsupervised tasks
2	Backpressure & Overload	load-	"The BEAM handles load" — cast on externally-driven ingest, push pipelines with no demand signal, :infinity/blind-retry timeout handling, input-sized Task.async fan-out
3	Event Delivery Semantics	evt-	"The event will arrive, once, in order" — PubSub as sole carrier of business facts, side effects enqueued outside the creating transaction, non-idempotent consumers, cross-source ordering, blocking/detaching telemetry handlers, persisted broadcast snapshots
4	Shared State & Consistency	state-	"I just checked, it's still true" — lookup-then-write races on ETS/Registry, pure-read calls serializing hot paths, persistent_term.put at runtime cadence
5	Distribution Reality	dist-	"One per cluster" enforced by node-local tools — :global singletons without fencing, Registry/ETS uniqueness assumed cluster-wide
6	Runtime Mechanics	mech-	Misused clocks, binaries, atoms — wall-clock duration math, sub-binaries pinning large parents in long-lived state, atoms minted from external input

Gotchas

Read gotchas.md before dispatching reviewers — it pre-records scope guards (the one fail-open rule, the shapes-not-brands principle, telemetry attach-versus-emit, wall-clock-as-data, cast not being banned per se) so reviewers do not judge outside the rules.

Related Skills

• adversarial-elixir — the paradigm-fit sibling gate: OO/enterprise habits ported onto the BEAM (layering, processes as objects, anemic data, defensive flow). Same protocol, complementary rules; run both on substantial OTP work.
• staff-level-elixir — the advisory sibling: greenfield "which tool, which convention" guidance and judgment calls both gates deliberately exclude. Use it to write or fix code; use the gates to verdict it.
• elixir-meta-programming — how to build macros/DSLs correctly when one is warranted; out of this gate's scope entirely.

Reference Files

File	Description
references/reviewer-prompt.md	Self-contained prompt template for each blind reviewer
assets/templates/verdict.md	Verdict report template
references/_sections.md	Category definitions and ordering
metadata.json	Version and source references