Other formats agents might prefer:
markdownjsonllms.txt

Agent? You probably want markdown or json, or Pod over MCP.

Reported issues for ssh-mcp-pro

Pod holds 7 of 7 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to ssh-mcp-pro.

Most discussed

hardening(remote): add WebSocket lifecycle limits and standards-compliant framing

Summary

Harden the remote agent WebSocket boundary with bounded pre-authentication lifetime, connection limits, and a maintained standards-compliant implementation.

Background

The current minimal WebSocket implementation limits frame bytes but does not provide the breadth of protocol validation and lifecycle controls expected for an internet-facing control plane. Pre-authentication connections should not remain open indefinitely or consume unbounded resources.

Scope

  • Add an agent-

Read the thread · 2026-07-22 · closed · 0 comments

test(http): cover real HTTP server authentication and session boundaries

Summary

Add in-process tests for the actual HTTP server entrypoint and include security-relevant server wiring in meaningful coverage reporting.

Background

Helper-level HTTP security tests are strong, but src/server-http.ts and src/render-http.ts are excluded from coverage. The real boundary combines CORS, bearer/OAuth authorization, session lifecycle, body parsing, transport setup, rate-limit headers, and graceful shutdown.

Scope

  • Create an isolated in-process HTTP test harne

Read the thread · 2026-07-22 · closed · 0 comments

deps: remediate production advisories and restore a clean audit gate

Summary

Upgrade, override, or otherwise remediate currently reported production dependency advisories so the configured audit gate passes without suppressing actionable findings.

Background

The audit snapshot reported production-path findings in OpenTelemetry, fast-uri, protobufjs, Hono, and the Hono Node adapter. Development-only findings also exist, but runtime dependencies take priority because they block pnpm run check and affect published artifacts.

Scope

  • Update direct

Read the thread · 2026-07-22 · closed · 0 comments

security(agent): add replay protection and monotonic validation for policy updates

Summary

Apply nonce, freshness, and strictly monotonic version validation to signed policy.update envelopes on the agent.

Background

Action requests have replay and deadline checks, but policy updates are accepted after signature validation without equivalent nonce-window, timestamp-age, or version monotonicity enforcement. A previously signed policy must not be reusable to roll an agent back.

Scope

  • Track recently accepted policy-update nonces with a bounded TTL window.
  • Rejec

Read the thread · 2026-07-22 · closed · 0 comments

security(http): validate the effective remote connector tool set before public binding

Summary

Base non-loopback HTTP safety decisions on the effective exposed tool set, not only on the configured profile name.

Background

Client-specific extra-tool environment variables can extend connector profiles. The current startup validation considers every non-full profile remote-safe even when extensions add tools that are outside the approved remote connector surface.

Scope

  • Introduce a canonical immutable allowlist for tools permitted on remote-safe profiles.
  • Reject un

Read the thread · 2026-07-22 · closed · 0 comments

security(agent): enforce tool-to-capability authorization invariants

Summary

Enforce the canonical mapping between each remote tool and its required capability inside the agent before dispatching any action.

Background

The control plane currently emits the expected tool/capability pair, but the agent must remain an independent authorization boundary. A controlled audit demonstrated that trusting the capability field without re-deriving it from the tool can permit an action under an unrelated enabled capability.

Scope

  • Resolve the expected capabilit

Read the thread · 2026-07-22 · closed · 0 comments

security(agent): canonicalize local paths and prevent symlink boundary escapes

Summary

Make local file and log authorization operate on canonical filesystem targets and fail closed when symlinks or path races cross policy boundaries.

Background

The current remote-agent path policy performs lexical prefix checks before filesystem operations that may follow symlinks. A controlled audit confirmed that a lexically allowed path can resolve to a target outside the configured allowlist.

Scope

  • Canonicalize existing read targets with realpath before policy evaluat

Read the thread · 2026-07-22 · closed · 0 comments

Most recent

The remaining reports are on the project's issue tracker.