AI Agent Workflow Collection

Expert review pass for data-flow / state-ownership / lifetime / boundary topology -- the lens that asks "who creates this, who owns it, who consumes it, who decides about it" at every non-trivial entity in scope, and flags mismatches between conceptual scope and instantiation scope (singleton inside a per-note constructor), wrong direction of data flow (consumer reaches up into producer), constructors that reach out into the world, hidden state via module-mutable / closure capture, identity-vs-value confusion (Hickey), wrong dependency direction (domain depending on infrastructure), missing teardown on owned objects, and boundary placement that doesn't match the natural seams. The "draw the data-flow diagram before reading the body" discipline. Especially load-bearing in a world where the implementation body is increasingly AI-written and routinely regenerated -- the topology survives replacements; the body doesn't. Reviews the current branch diff against main by default, a specific file/PR with /data-flow-review or /data-flow-review , or a git range. Auto-delegates depth to the data-flow subagent. Produces severity-labeled findings with file:line references plus a data-flow diagram per finding when the alternative is concrete. Does NOT post comments.

Expert review pass for distributed-systems footguns in changed code -- retries without idempotency, unbounded queues, fire-and-forget calls without dedup, missing fencing tokens, cache stampedes, replication-lag races, and other patterns that pass tsc/rustc/pytest but produce 3am pages. Reviews the current branch diff against main by default, a specific file/PR with /distsys-review or /distsys-review , or a git range. Auto-routes deep questions to distsys-data or distsys-runtime subagents. Produces severity-labeled findings with file:line references. Does NOT post comments. Use when reviewing changes that touch persistence, messaging, retries, caching, or anything that crosses a process boundary.