Welcome: share reproducible MCP reliability evidence

[aliengineering-byte]

Welcome to ResiliReplay Discussions.

Use this space for bounded, technical questions about MCP reliability testing, campaign design, recovery expectations, and sanitized field evidence. A useful post includes:

• the exact ResiliReplay and server versions;
• the transport and one reviewed target/tool boundary;
• declared expectations separated from actual results;
• reproduction commands, hashes, and a compact report;
• cleanup performed and limitations.

Start with the five-minute field-test guide and compare the three reproducible external cases.

Do not post credentials, private paths, production traces, personal data, browser profiles, or destructive proofs of concept. Synthetic injected failures are reliability conditions, not vulnerability claims. Use private security reporting for a possible vulnerability.

Independent users may propose a truthful adopter entry only with a public reproducible reference. Maintainer-run case studies are not adopter claims.

[aliengineering-byte]

v0.5.0 “Everywhere” is public

ResiliReplay can now turn supported Claude Code, Codex, and Hermes tool-failure events into one canonical, sanitized evidence format and an executable regression—without enabling capture at install time.

Try the public package:

npx --yes resilireplay@0.5.0 demo
npx --yes resilireplay@0.5.0 connect --agent auto --dry-run
npx --yes resilireplay@0.5.0 mcp serve

What changed:

• four versioned public schemas and one canonical @resilireplay/agent integration core;
• opt-in passive capture with bounded files/events, redaction, atomic writes, corruption recovery, deduplication, and path-containment checks;
• portable Agent Skill plus real Claude and Codex plugin packages;
• Hermes skill discovery and local stdio MCP registration;
• a nine-tool MCP server validated with official MCP Inspector 2.1.0;
• deterministic adapter init/verify and conformance fixtures;
• a self-contained npm tarball with SLSA provenance.

The release gate passed 91 tests across 16 files, Ubuntu/Windows on Node 22/24, Studio Playwright E2E, packed-package smoke, secret/privacy scans, and the official Claude plugin and Agent Skills validators.

Evidence labels remain intentionally narrow: Claude and Codex are INSTALLATION + FIXTURE VERIFIED (not claimed as live model flows); Hermes is INSTALLATION VERIFIED; future adapters remain documented-only until their conformance and runtime paths are exercised.

• Release
• npm 0.5.0
• Compatibility matrix
• Release evidence

If you maintain an agent or MCP server, a useful external field report is one sanitized, reproducible failure boundary plus the expected recovery invariant—never credentials, production traces, or private paths.