feat: implement PLAN_NEXT wave 2 agent detection/correlation/policy#89
feat: implement PLAN_NEXT wave 2 agent detection/correlation/policy#89
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 261abf233b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } | ||
| } | ||
|
|
||
| agentContext := agentContextByID[tool.AgentID] |
There was a problem hiding this comment.
Build looks up agent context with agentContextByID[tool.AgentID], but tool.AgentID is derived from identity.AgentID(toolID, org) while agent records are keyed by identity.AgentID(instanceID, org) in inventory construction (core/aggregate/inventory/inventory.go uses toolID at lines 218-223 and instanceID at lines 242-254). For framework findings that include a symbol/range, these IDs diverge, so bound/deployment context is dropped from agent_privilege_map even when inventory.agents has it.
Useful? React with 👍 / 👎.
| if len(agents) == 0 { | ||
| return legacyRuleResult("block_secret_presence", findings) | ||
| } |
There was a problem hiding this comment.
This branch only falls back to block_secret_presence when there are zero agent findings, so as soon as any agent is detected, WRKR-A002 stops evaluating secret_presence entirely and only checks human-gate behavior. Because rule IDs are canonicalized to the WRKR-### family in core/policy/loader.go (lines 84-89), this effectively changes WRKR-002 behavior in mixed repos and can let leaked-secret findings pass unnoticed.
Useful? React with 👍 / 👎.
| gate := strings.ToLower(strings.TrimSpace(evidenceValue(finding, "deployment_gate"))) | ||
| if gate != "approved" && gate != "enforced" { |
There was a problem hiding this comment.
The rule requires deployment_gate evidence for every auto_deploy=true agent, but this commit introduces no detector/emitter path for that key (repo search shows deployment_gate only in this evaluator and its test). In practice, auto-deploy agents will always violate WRKR-A009 regardless of configuration, creating unavoidable policy failures and noisy results.
Useful? React with 👍 / 👎.
1 participant
Problem
Wave 2 work from
product/PLAN_NEXT.mdneeded implementation for:Changes
agentlangchain,agentcrewai,agentopenai,agentautogen,agentllamaindexcore/aggregate/agentresolvercore/aggregate/agentdeployagent_privilege_mapentries additively with bindings/deployment/approval context.Validation
make prepush-fullgo run ./cmd/wrkr scan --path scenarios/wrkr/scan-mixed-org/repos --json --quietgo test ./core/detect/... -count=1go test ./core/aggregate/... ./core/detect/compiledaction ./core/cli -count=1go test ./core/policy/... -count=1make test-contractsmake test-scenariosmake test-hardeningmake test-chaos