@aryaminus/controlkeel-opencode 0.2.46 → 0.2.49

package/.opencode/agents/controlkeel-operator.md

@@ -22,6 +22,7 @@ and validate them against the project's security, budget, and compliance policie
 
 - `ck_context` — Load mission, findings, budget, and proof context
 - `ck_validate` — Run full governance validation
+- `ck_execute_code` — Run generated code only through CK's guarded Docker sandbox; use `dry_run` first and do not request network/secrets/shell/deploy
 - `ck_finding` — Record a governed finding when you detect a missed issue
 - `ck_review_submit` — Submit review material for human approval
 - `ck_review_status` — Check review status before execution

package/.opencode/plugins/controlkeel-governance.ts

@@ -360,24 +360,27 @@ export const ControlKeelGovernance: Plugin = async ({ project, client, $, direct
       const openError = typeof openPayload?.open_error === "string" ? openPayload.open_error.trim() : ""
       const openFailure = typeof openPayload?.error === "string" ? openPayload.error.trim() : ""
       const browserNotOpened = openPayload?.opened !== true
+      const serverUnavailable = openPayload?.server_serving === false
 
       const remoteLocalhostMismatch =
         typeof browserUrl === "string" &&
         browserUrl.includes("localhost") &&
         openPayload?.remote === true
 
-      if (!browserUrl || openError || openFailure || remoteLocalhostMismatch || browserNotOpened) {
+      if (!browserUrl || serverUnavailable || openError || openFailure || remoteLocalhostMismatch || browserNotOpened) {
         return buildPlanResult({
           waitSkipped: true,
           manualApprovalRequired: true,
           reason:
             !browserUrl
               ? "browser_url_unavailable"
-              : browserNotOpened
-                ? "browser_not_opened"
-                : "browser_unreachable",
+              : serverUnavailable
+                ? "review_server_unavailable"
+                : browserNotOpened
+                  ? "browser_not_opened"
+                  : "browser_unreachable",
           guidance:
-            "Browser review is unavailable from this environment or did not actually open. Ask the user for explicit approval in chat, then record it with `controlkeel review plan respond --id <review_id> --decision approved --feedback-notes \"User approved in chat; browser unavailable\" --json` or `ck_review_feedback`.",
+            "Browser review is unavailable, the CK review server is not reachable, or the browser did not actually open. Ask the user for explicit approval in chat, then record it with `controlkeel review plan respond --id <review_id> --decision approved --feedback-notes \"User approved in chat; browser/review server unavailable\" --json` or `ck_review_feedback`.",
         })
       }
 

package/.opencode/skills/controlkeel-governance/SKILL.md

@@ -13,6 +13,7 @@ compatibility:
   - open-standard
 allowed-tools:
   - ck_validate
+  - ck_execute_code
   - ck_context
   - ck_finding
   - ck_memory_search
@@ -32,6 +33,7 @@ metadata:
   category: governance
   ck_mcp_tools:
     - ck_validate
+    - ck_execute_code
     - ck_context
     - ck_finding
     - ck_memory_search
@@ -53,16 +55,17 @@ You are operating inside a **ControlKeel-governed session**. Start here whenever
 
 1. Call `ck_context` at task start to load mission, risk, budget, proof, active findings, workspace context, context reacquisition, instruction hierarchy, and recent transcript state.
 2. Call `ck_validate` before writing code, config, shell, or deploy text, and pass trust-boundary metadata when the source content came from the web, tools, skills, or mixed provenance.
-3. If you discover a problem the scanner did not raise, call `ck_finding`.
-4. Use `ck_memory_search` when you need explicit recall of prior decisions, checkpoints, or findings rather than relying only on the default context packet.
-5. Use `ck_memory_record` to persist important decisions, assumptions, and operator guidance that future agents should recover.
-6. Use `ck_memory_archive` to retire stale or superseded guidance before it keeps contaminating retrieval.
-7. Call `ck_budget` and `ck_cost_optimizer` before expensive model or bulk operations.
-8. Call `ck_route` before delegating sub-work to another agent.
-9. Use `ck_deployment_advisor` to analyze stack and generate deployment templates when checking ship readiness.
-10. Use `ck_regression_result` to record external browser or QA evidence before claiming deploy readiness.
-11. Use `ck_outcome_tracker` to track success/failure outcomes for continuous learning.
-12. Use `ck_skill_list` and `ck_skill_load` to activate more specific CK workflows.
+3. Use `ck_execute_code` only for generated code that should run inside CK's guarded Docker sandbox; prefer `dry_run` first, and never treat it as local shell access or a network/secrets grant.
+4. If you discover a problem the scanner did not raise, call `ck_finding`.
+5. Use `ck_memory_search` when you need explicit recall of prior decisions, checkpoints, or findings rather than relying only on the default context packet.
+6. Use `ck_memory_record` to persist important decisions, assumptions, and operator guidance that future agents should recover.
+7. Use `ck_memory_archive` to retire stale or superseded guidance before it keeps contaminating retrieval.
+8. Call `ck_budget` and `ck_cost_optimizer` before expensive model or bulk operations.
+9. Call `ck_route` before delegating sub-work to another agent.
+10. Use `ck_deployment_advisor` to analyze stack and generate deployment templates when checking ship readiness.
+11. Use `ck_regression_result` to record external browser or QA evidence before claiming deploy readiness.
+12. Use `ck_outcome_tracker` to track success/failure outcomes for continuous learning.
+13. Use `ck_skill_list` and `ck_skill_load` to activate more specific CK workflows.
 
 ## Non-negotiable rules
 
@@ -81,6 +84,7 @@ You are operating inside a **ControlKeel-governed session**. Start here whenever
 
 - `ck_context` — mission, task, budget, proof, memory, workspace snapshot, transcript summary, resume context
 - `ck_validate` — governed preflight scan with trust-boundary checks
+- `ck_execute_code` — guarded generated-code execution; Docker sandbox only, local/network/secrets/shell/deploy denied, `dry_run` recommended first
 - `ck_finding` — persist manual findings
 - `ck_memory_search`, `ck_memory_record`, `ck_memory_archive` — explicit typed-memory retrieval and hygiene
 - `ck_regression_result` — import external regression evidence into proof state

package/AGENTS.md

@@ -11,8 +11,9 @@ Required workflow:
 2. Call `ck_validate` before writing code, config, shell, or deploy content.
 3. Submit plans or approval packets with `ck_review_submit` and check `ck_review_status` before execution.
 4. Record any human-review issue with `ck_finding`.
-5. Check `ck_budget` before expensive model or multi-agent work.
-6. Use `ck_route`, `ck_skill_list`, and `ck_skill_load` to delegate or activate specialized CK workflows.
+5. Check `ck_budget` before expensive model or multi-agent work, and keep `ck_context` compact unless full raw context is needed.
+6. Before AFK or delegated implementation, split large work into human-approved vertical slices with explicit dependencies; prefer durable behavior-first issues, stable deep-module interfaces, and branch-level automated review plus human QA before merge.
+7. Use `ck_route`, `ck_skill_list`, and `ck_skill_load` to delegate or activate specialized CK workflows.
 
 Install ControlKeel:
 - Homebrew: `brew tap aryaminus/controlkeel && brew install controlkeel`

package/index.js

@@ -343,24 +343,27 @@ export const ControlKeelGovernance = async ({ $, directory }) => {
       const openError = typeof openPayload?.open_error === "string" ? openPayload.open_error.trim() : ""
       const openFailure = typeof openPayload?.error === "string" ? openPayload.error.trim() : ""
       const browserNotOpened = openPayload?.opened !== true
+      const serverUnavailable = openPayload?.server_serving === false
 
       const remoteLocalhostMismatch =
         typeof browserUrl === "string" &&
         browserUrl.includes("localhost") &&
         openPayload?.remote === true
 
-      if (!browserUrl || openError || openFailure || remoteLocalhostMismatch || browserNotOpened) {
+      if (!browserUrl || serverUnavailable || openError || openFailure || remoteLocalhostMismatch || browserNotOpened) {
         return buildPlanResult({
           waitSkipped: true,
           manualApprovalRequired: true,
           reason:
             !browserUrl
               ? "browser_url_unavailable"
-              : browserNotOpened
-                ? "browser_not_opened"
-                : "browser_unreachable",
+              : serverUnavailable
+                ? "review_server_unavailable"
+                : browserNotOpened
+                  ? "browser_not_opened"
+                  : "browser_unreachable",
           guidance:
-            "Browser review is unavailable from this environment or did not actually open. Ask the user for explicit approval in chat, then record it with `controlkeel review plan respond --id <review_id> --decision approved --feedback-notes \"User approved in chat; browser unavailable\" --json` or `ck_review_feedback`.",
+            "Browser review is unavailable, the CK review server is not reachable, or the browser did not actually open. Ask the user for explicit approval in chat, then record it with `controlkeel review plan respond --id <review_id> --decision approved --feedback-notes \"User approved in chat; browser/review server unavailable\" --json` or `ck_review_feedback`.",
         })
       }
 

package/package.json

@@ -35,5 +35,5 @@
     "url": "git+https://github.com/aryaminus/controlkeel.git"
   },
   "type": "module",
-  "version": "0.2.46"
+  "version": "0.2.49"
 }