feat(agent): drop tool allowlists and turn caps, widen audit protocols, bump v1.45.0 - #7
Merged
Merged
Conversation
…s, bump v1.45.0 Subagent tools frontmatter accepts tool names only; Bash(cmd *) entries resolved to bare Bash and never restricted anything, while the prompts written around them made agents avoid rg/grep/Edit they actually had. Remove the allowlists from all 15 agents so specialization comes from the prompt, drop maxTurns on rust-code-reviewer and rust-critic, make the startup Skill() call best-effort with a fallback to preloaded skills, and move the security and architecture analysts to opus. security-audit gains an attack surface map, network and filesystem hardening, verification gates, and the missing vectors (SSRF, TLS, XXE, ReDoS, zip-slip, TOCTOU, async starvation, registry sources, cargo vet). arch-inspect gains error-handling design, API stability, observability, lint and manifest hygiene, and a toolchain evidence step. rust-security-maintenance now preloads security-audit. Fix contradictions in rust-live-tester and rust-researcher write scopes, the duplicate rust-modern-apis call in rust-code-reviewer, and the tech-writer model mismatch between README and definition.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Removes the restrictions that limited the agents beyond their specialization and widens the security and architecture audit protocols.
toolsallowlists removed. Subagenttoolsaccepts tool names only;Bash(cmd *)entries resolved to bareBash(verified with a probe agent:echo,date,mkdir,RUST_LOG=... cargoall ran under aBash(cargo *)list) and never restricted anything, while the prompts written around them told agents they had norg/grep/Edit. Specialization now lives in the prompt.maxTurnsremoved fromrust-code-reviewer(20) andrust-critic(15): full reviews came back as partial.Skill()calls are best-effort with a fallback to theskills:preload (some hosts spawn plugin subagents without theSkilltool).rust-security-analystandrust-arch-analystonclaude-opus-5;tech-writerfromhaikutosonnet.security-audit: attack surface map, network and filesystem hardening, verification gates, and the missing vectors (SSRF, TLS verification, XXE, ReDoS, zip-slip, decompression bombs, TOCTOU, async starvation, cancellation safety, registry sources, gitrevpinning,cargo vet, typosquatting).rust-security-maintenancenow preloads it so the fixer and the auditor share one catalogue.arch-inspect: toolchain evidence step, error-handling design, API stability, observability, lint and manifest hygiene, async section extended..local/testing/-only write scopes vs the mandatory handoff, researcher spawningsddwithoutAgent, duplicaterust-modern-apiscall in the reviewer, README model mismatch fortech-writer.Full analysis behind the change:
.local/reports/2026-09-08-agent-restrictions-review.md(local, not committed).Verification
claude plugin validate --strict --json ./rust-codeand.: no errors, no warnings.1.45.0.tools,maxTurns, orpermissionModelines remain in any agent frontmatter.