protect-agent-config
Guard against an agent hand-editing its own guardrails. Agent instruction files (AGENTS.md and the per-agent wrappers), permission files (.claude/settings.json, .mcp.json), the dependency allowlist (.chock/dependency-allowlist.txt) and vendored enforcement (.chock/bin/, .chock/compiled/) define what the agent may do -- so a shell command that rewrites them is the agent modifying its own authority (MITRE ATLAS AML.T0081). The guard refuses shell writes to those paths -- a redirect, rm/mv/tee/sed -i, cp into the path, git checkout/restore, PowerShell Set-Content/Add-Content/Out-File; reads and copies out pass, and `chock sync` passes. Best-effort and deliberately coarse. The 'chock: approved-config-change' marker is friction plus an audit trail, not authentication. A second, tool_use-only gate refuses Edit/Write to the same paths; it never runs at commit, so a person stays free to edit them.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 10
- Forks
- 1
01 前言 Front Matter
- name
- protect-agent-config
- metadata
- { "chock.artifact": "rule", "chock.enforcement": "advise", "chock.coverage_without_chock": "advisory" }
- description
- Guard against an agent hand-editing its own guardrails. Agent instruction files (AGENTS.md and the per-agent wrappers), permission files (.claude/settings.json, .mcp.json), the dependency allowlist (.chock/dependency-allowlist.txt) and vendored enforcement (.chock/bin/, .chock/compiled/) define what the agent may do -- so a shell command that rewrites them is the agent modifying its own authority (MITRE ATLAS AML.T0081). The guard refuses shell writes to those paths -- a redirect, rm/mv/tee/sed -i, cp into the path, git checkout/restore, PowerShell Set-Content/Add-Content/Out-File; reads and copies out pass, and `chock sync` passes. Best-effort and deliberately coarse. The 'chock: approved-config-change' marker is friction plus an audit trail, not authentication. A second, tool_use-only gate refuses Edit/Write to the same paths; it never runs at commit, so a person stays free to edit them.
03 SKILL.md 正文
Protect Agent Config
Guard against an agent hand-editing its own guardrails. Agent instruction files (AGENTS.md and the per-agent wrappers), permission files (.claude/settings.json, .mcp.json), the dependency allowlist (.chock/dependency-allowlist.txt) and vendored enforcement (.chock/bin/, .chock/compiled/) define what the agent may do -- so a shell command that rewrites them is the agent modifying its own authority (MITRE ATLAS AML.T0081). The guard refuses shell writes to those paths -- a redirect, rm/mv/tee/sed -i, cp into the path, git checkout/restore, PowerShell Set-Content/Add-Content/Out-File; reads and copies out pass, and chock sync passes. Best-effort and deliberately coarse. The 'chock: approved-config-change' marker is friction plus an audit trail, not authentication. A second, tool_use-only gate refuses Edit/Write to the same paths; it never runs at commit, so a person stays free to edit them.
agent_config(AGENTS.md|wrappers|.claude/settings|.mcp.json|.chock/dependency-allowlist.txt|.chock/bin|.chock/compiled|.agents/policies/*/implementations): never(hand_edit|delete); regenerate_via(chock sync)
if(config_change_needed): propose_to_human; await(approval) # an agent must not widen or disarm its own guardrails
This skill is advisory: the client reading it has no mechanism to enforce it. The same policy compiled by chock can refuse an agent's shell command before it runs; blocks on an agent's file writes and at turn end. See https://github.com/open-coder-ai/chock
04 同仓库更多 Skills
validate
Lint Chock policy conformance. args(policy_id or all) returns(findings, verdict) invoke(validate, check, promotion_to_review) exclude(eval, optimize)
policy-init
Create conformant Chock policy from request. args(request, target_path, artifact_hint) returns(folder, wiring) invoke(skill, hook, rule, workflow, convention, always_never) exclude(coding, edit_existing)
optimize
Improve Chock policy from usage evidence. args(policy_path, >=3 traces) returns(promoted_edit / rejected_record / no_bar) invoke(misfire, ignore, tuning) exclude(policy_init, one_off_edit)
eval
Run Chock policy eval suite. args(policy_path) returns(pass_rate, verdict) invoke(test, run_evals, promotion_check) exclude(validate, optimize)
chock-init
Onboard a repo into Chock. args(repo_path, agents, agent_agnostic) returns(wiring) invoke(onboard, scaffold, setup) exclude(coding, policy_creation)
validate
Lint Chock policy conformance. args(policy_id or all) returns(findings, verdict) invoke(validate, check, promotion_to_review) exclude(eval, optimize)