name: exeris-subsystem-specialist description: 'Subsystem-focused PR review for Exeris Kernel using one skill with modes: bootstrap, memory, crypto, transport, persistence, events, flow, http, security, config. Use when you want a targeted subsystem review without creating many separate agents.' argument-hint: 'Mode (bootstrap|memory|crypto|transport|persistence|events|flow|http|security|config) + PR scope/files + review goal' user-invocable: true disable-model-invocation: false
Exeris Subsystem Specialist
Purpose
Provide one lightweight, mode-driven subsystem review skill instead of many dedicated agents.
This skill keeps one shared Exeris philosophy while switching to subsystem-specific checks via mode.
Supported Modes
- bootstrap
- memory
- crypto
- transport
- persistence
- events
- flow
- http
- security
- config
Invocation Pattern
Use a mode-first subprompt, for example:
- Review this as Transport subsystem
- Review this as Memory subsystem
- Review this as Security subsystem
If mode is not explicit, infer from changed files. If still ambiguous, ask for one mode before continuing.
Canon to Load Before Review
Always load:
- docs/architecture.md
- docs/whitepaper.md
- docs/performance-contract.md
- docs/modules/01-spi.md
- docs/modules/02-core.md
- docs/modules/03-community.md
- docs/modules/04-enterprise.md
- docs/modules/05-tck.md
- related ADRs in docs/adr
Then load exactly one subsystem contract based on selected mode:
- bootstrap -> docs/subsystems/bootstrap.md
- memory -> docs/subsystems/memory.md
- crypto -> docs/subsystems/crypto.md
- transport -> docs/subsystems/transport.md
- persistence -> docs/subsystems/persistence.md
- events -> docs/subsystems/events.md
- flow -> docs/subsystems/flow.md
- http -> docs/subsystems/http.md
- security -> docs/subsystems/security.md
- config -> docs/subsystems/config.md
Mode Workflow
Select mode
- Explicit mode from user wins.
- Otherwise infer from changed symbols/files.
- If multiple modes compete, choose primary mode and list secondary impacts.
Build review lens for selected mode
- Extract contract invariants, lifecycle semantics, performance limits, and error model from subsystem doc.
- Extract cross-module boundaries from module docs and ADRs.
Run focused checks
- Contract correctness: changed behavior must match subsystem contract semantics.
- Boundary integrity: no SPI leakage, no Core driver leakage, no The Wall breach.
- Performance law: no regressions against No Waste Compute constraints.
- TCK implications: contract changes require executable TCK impact analysis and bindings.
Run mode-specific deep checks
- bootstrap: startup lifecycle ordering, bootstrap events, readiness/latency gates.
- memory: allocator ownership, leak safety, ref-count discipline, O(1) alloc-release behavior.
- crypto: handshake lifecycle, zero-alloc constraints, native boundary hygiene, error-code contracts.
- transport: connection/stream lifecycle, backpressure/load-shed semantics, zero-copy data path.
- persistence: transaction boundaries, rollback semantics, concurrency invariants, consistency contract.
- events: delivery semantics, ordering/idempotency expectations, provider-switching contract behavior.
- flow: scheduling and park-wake semantics, latency/throughput invariants, state-machine correctness.
- http: provider/server/client/handler/exchange contract conformance at SPI boundary.
- security: context propagation, identity/authorization invariants, isolation and auditability guarantees.
- config: schema and validation semantics, deterministic resolution/override precedence, boot-time safety.
Decision and report
- Output verdict: APPROVE, CONDITIONAL, REJECT.
- Map findings to mode contract clause, impacted symbols, runtime risk, and minimal corrective action.
Decision Logic
- APPROVE: selected mode contract is fully preserved and no cross-cutting guardrail is violated.
- CONDITIONAL: localized drift with clear remediation and no fundamental contract break.
- REJECT: subsystem contract violation, boundary breach, performance law breach, or missing TCK implications for contract changes.
Completion Criteria
A review is complete only if all are true:
- Mode selection is explicit and justified.
- Selected subsystem doc was loaded and used as primary contract source.
- Cross-cutting checks were run (architecture, performance, TCK).
- Mode-specific deep checks were run and evidenced.
- Verdict and required actions are precise and minimal.
Review Output Template
- Scope analyzed (mode, files, symbols)
- Contract findings (selected subsystem)
- Boundary findings (SPI/Core/The Wall)
- Performance findings (No Waste Compute)
- TCK findings (contract test implications)
- Verdict (APPROVE | CONDITIONAL | REJECT)
- Required actions (minimal root-cause fixes)
Non-Negotiable Rules
- One review, one primary mode, one contract source of truth.
- No subsystem review without loading its subsystem doc.
- No contract-changing PR accepted without TCK implications analysis.
- No violation of architecture boundaries or performance contract under any mode.