aiblueprint-cli 1.4.99 → 1.4.100

package/agents-config/skills/apex/steps/step-10-verify.md

@@ -1,296 +1,78 @@
 ---
 name: step-10-verify
-description: Launch app and verify feature works through real user surface
-prev_step: steps/step-04-validate.md
-next_step: steps/step-09-finish.md
+description: Prove required APEX acceptance criteria through the real user, API, provider, artifact, or deployment surface with current evidence.
+next_step: step-09-finish.md
 ---
 
-# Step 10: Verify (Feature Validation)
+# Step 10: Runtime proof
 
-## MANDATORY EXECUTION RULES (READ FIRST):
+Runtime proof is a hard gate when requested by the user, required by project rules, or selected by risk. Tests and code inspection support proof but do not replace a stronger required surface.
 
-- 🛑 NEVER skip verification - that defeats the purpose
-- 🛑 NEVER claim feature works without actually testing it
-- 📐 ALWAYS read the project's recommended verification rules FIRST (`.agents/rules/`, `AGENTS.md`, `CLAUDE.md`) and follow them
-- ✅ ALWAYS launch a verifier agent to test through real surface
-- ✅ ALWAYS compare results against original user request
-- 📸 ALWAYS capture 1+ screenshots proving the feature works - verification is INCOMPLETE without visual proof
-- ✅ ALWAYS fix issues before proceeding (unless user skips)
-- 📋 YOU ARE A VERIFIER, not an implementer
-- 💬 FOCUS on "Does the feature actually work for a real user?"
-- 🚫 FORBIDDEN to proceed with FAIL verdict without user confirmation
+## 1. Set proof state
 
-## EXECUTION PROTOCOLS:
+Set `{proof_gate}=NOT_PROVEN`. Record environment, revision, target surface, authentication state, fixture identity, and evidence directory.
 
-- 🎯 Launch verifier agent with full context
-- 💾 Log verification results (if save_mode)
-- 📖 Compare each AC against real behavior
-- 🚫 FORBIDDEN to fake verification
+Read project verification rules and use the approved local server, browser, simulator, CLI, API, provider, or release workflow. Reuse healthy managed services instead of starting duplicates.
 
-## CONTEXT BOUNDARIES:
+## 2. Build the proof matrix
 
-- Implementation is complete, validated, and reviewed
-- All typecheck/lint/tests pass
-- Now testing the REAL user experience
-- **If `{teams_mode}` = true:** Agent team is still alive. Do NOT shutdown teammates - that happens in step-09-finish only.
+Create one row per observable contract:
 
-## YOUR TASK:
+| ID | Acceptance criterion | Starting state | Action | Expected result | Evidence layer | Artifact | Status |
+|---|---|---|---|---|---|---|---|
 
-Launch the application, test the feature through real user interaction, and verify it matches the original request.
+Include the initial state, meaningful transitions, final outcome, and relevant negative, persistence, refresh, permission, or regression paths implied by the request.
 
----
-
-<available_state>
-From previous steps:
-
-| Variable | Description |
-|----------|-------------|
-| `{task_description}` | Original user request |
-| `{task_id}` | Kebab-case identifier |
-| `{acceptance_criteria}` | Success criteria from analysis |
-| `{auto_mode}` | Skip confirmations |
-| `{save_mode}` | Save outputs to files |
-| `{economy_mode}` | No subagents |
-| `{output_dir}` | Path to output (if save_mode) |
-| Files modified | From step-03 |
-</available_state>
-
----
-
-## EXECUTION SEQUENCE:
-
-### 1. Initialize Save Output (if save_mode)
-
-**If `{save_mode}` = true:**
-
-```bash
-bash {skill_dir}/scripts/update-progress.sh "{task_id}" "10" "verify" "in_progress"
-```
-
-Append results to `{output_dir}/10-verify.md` as you work.
-
-### 2. Gather Context for Verifier
-
-Collect all information the verifier agent needs:
-
-```
-1. Original task description: {task_description}
-2. Acceptance criteria: {acceptance_criteria}
-3. Modified files (from git):
-   git diff --name-only HEAD~1
-   git status --porcelain
-4. Implementation summary from step-03/04
-```
-
-### 2b. Load Recommended Verification Rules
-
-Look for project-specific verification conventions and **follow them**. They override the defaults below. Check in order:
-
-1. `.agents/rules/` — any rule file about verification, testing, QA, or e2e (e.g. `verify.md`, `testing.md`)
-2. `AGENTS.md` — verification / commands sections
-3. `CLAUDE.md` — verification preferences (e.g. "Prefer /dev-browser for verification")
-4. `README.md` + `package.json` scripts — how to launch / test
-
-```bash
-ls .agents/rules/ 2>/dev/null
-cat AGENTS.md CLAUDE.md 2>/dev/null
-```
-
-Capture the recommended launch command, auth/login steps, test URLs/credentials, and any required evidence into `{verification_rules}` and pass it to the verifier. If nothing is documented, fall back to detecting the dev/start command from `package.json`.
-
-### 3. Launch Verifier Agent
-
-**If `{economy_mode}` = true:**
--> Self-verify: manually test the feature using available tools (dev-browser, curl, shell commands). FOLLOW the recommended verification rules from step 2b, capture 1+ screenshots as proof, and include the screenshots directly in the chat using Markdown image syntax with absolute local paths. Follow the verification process below directly instead of launching an agent.
-
-**If `{economy_mode}` = false:**
--> Launch a verifier sub-agent:
-
-```
-Sub-agent:
-  profile/type: "verifier"
-  prompt: |
-    ## Feature Verification
-
-    **Original Request:** {task_description}
-
-    **Acceptance Criteria:**
-    {acceptance_criteria - list each one}
-
-    **Files Modified:**
-    {list of modified files}
-
-    **Implementation Summary:**
-    {brief summary of what was done}
+Choose evidence that matches the surface:
 
-    **Recommended Verification Rules (FOLLOW THESE):**
-    {verification_rules - or "none documented; use defaults"}
+- visual step: current screenshot;
+- CLI/API: raw command or response artifact;
+- persistence: reload, relaunch, or authoritative state read-back;
+- provider: provider/API read-back, not local configuration alone;
+- public artifact/deployment: independently fetch the public surface or artifact;
+- authenticated live flow: controlled real interaction and final observable result.
 
-    Verify this feature works correctly by:
-    1. FOLLOW the recommended verification rules above (launch command, auth, URLs)
-    2. Launch the app (rules first, else package.json dev/start command)
-    3. Navigate to the relevant pages/endpoints
-    4. Test each acceptance criterion through real interaction
-    5. Use /dev-browser for web UI (capture screenshots), curl for APIs, shell for CLI tools
-    6. 📸 MANDATORY: Capture 1+ screenshots proving the finished feature works. For web UI use dev-browser saveScreenshot; for non-visual surfaces capture the terminal output / response as evidence
-    7. Output the screenshot(s) directly in the chat using Markdown image syntax with absolute local paths, e.g. `![Feature verification](/absolute/path/screenshot.png)`. Do NOT only list paths.
-    8. Report your findings with PASS/FAIL for each AC, listing the screenshot paths below the inline images
-    9. List anything missing compared to the original request
-```
-
-### 4. Process Verification Results
-
-Parse the verifier agent's report:
+## 3. Use an independent verifier when valuable
 
-**If no screenshot / evidence was captured or screenshots are only listed as paths without inline chat images:**
--> Verification is INCOMPLETE. Re-run capture (or send the verifier back) before accepting any PASS verdict.
+A fresh verifier context is useful for material user-facing, high-risk, or disputed flows. Give it the original request, acceptance criteria, verification rules, current revision, and proof matrix. The coordinator inspects every returned artifact before accepting it.
 
-**If all ACs PASS:**
--> Display success summary and proceed
+## 4. Exercise the real flow
 
-**If any ACs FAIL:**
--> Display failure details
+For each row:
 
-### 5. Present Verification Report
+1. Establish the documented starting state.
+2. Perform the action through the intended surface.
+3. Wait for and inspect the observable result.
+4. Check relevant errors, failed requests, crashes, logs, and persistent state.
+5. Capture evidence immediately with ordered artifact names.
+6. Record timestamp, environment, revision, action, observed result, and artifact path.
+7. Mark PASS only when the expected result is directly visible in current evidence.
 
-```markdown
-## Verification Results
+Do not reuse evidence invalidated by a later code, configuration, environment, or data change.
 
-**Feature:** {task_description}
+## 5. Evaluate and continue
 
-| AC | Status | Details |
-|----|--------|---------|
-| AC1 | ✓ PASS | Working as expected |
-| AC2 | ✗ FAIL | {what went wrong} |
+Set `{proof_gate}=PASS` only when all required criteria and rows pass, all artifacts exist and are current, and no observed error invalidates the flow.
 
-**Screenshots:** {inline Markdown image(s) REQUIRED, then list screenshot paths - at least 1}
-
-**Verdict:** {PASS or FAIL}
-```
+While the gate is not PASS:
 
-### 6. Handle Failures
+- identify the exact missing proof or failing behavior;
+- use the shortest real feedback loop to diagnose it;
+- return to planning or execution for in-scope fixes;
+- re-run affected validation and independent review;
+- reset the verification state and recapture every invalidated row.
 
-**If verdict = PASS:**
--> Proceed to next step
+There is no arbitrary retry limit while attempts produce meaningful progress. If a genuine external dependency blocks progress after safe alternatives are exhausted, report `BLOCKED — NOT PROVEN` with the exact condition and required input. Never relabel it as completion.
 
-**If verdict = FAIL:**
-
-**If `{auto_mode}` = true:**
--> Auto-fix: Apply fixes for each failing AC, then re-verify (max 2 retry rounds)
-
-**If `{auto_mode}` = false:**
-
-```yaml
-questions:
-  - header: "Verify"
-    question: "Verification found issues. How would you like to proceed?"
-    options:
-      - label: "Fix and re-verify (Recommended)"
-        description: "Apply fixes for failing ACs and verify again"
-      - label: "Fix without re-verify"
-        description: "Apply fixes but skip second verification"
-      - label: "Skip verification"
-        description: "Accept current state, proceed anyway"
-      - label: "Discuss issues"
-        description: "Let me review the findings first"
-    multiSelect: false
-```
+## 6. Present evidence
 
-### 7. Fix Loop (if user chooses to fix)
-
-```
-max_verify_rounds = 3
-round = 0
-
-WHILE round < max_verify_rounds AND verdict = FAIL:
-    round += 1
-
-    1. Read each failing AC
-    2. Identify the root cause
-    3. Apply the fix
-    4. Run typecheck + lint to ensure no regressions
-    5. Re-launch verifier agent (or self-verify if economy_mode)
-    6. Update verdict
-```
-
-**If still failing after 3 rounds:**
-
-```yaml
-questions:
-  - header: "Stuck"
-    question: "Verification still failing after {round} fix rounds. How proceed?"
-    options:
-      - label: "Keep trying"
-        description: "Continue fixing"
-      - label: "Accept current state"
-        description: "Proceed with known issues"
-      - label: "I'll fix manually"
-        description: "Let me investigate"
-    multiSelect: false
-```
+Show the proof matrix in flow order. Render visual artifacts inline with absolute local paths and link non-visual artifacts. State the exact local/static, provider, public-artifact/deployment, and authenticated-live boundaries proven.
 
-### 8. Complete Save Output (if save_mode)
-
-**If `{save_mode}` = true:**
-
-Append to `{output_dir}/10-verify.md`:
-```markdown
----
-## Step Complete
-**Status:** ✓ Complete
-**Verdict:** {PASS/FAIL}
-**Rounds:** {count}
-**Screenshots:** {inline Markdown image(s) first, then paths - at least 1}
-**Failing ACs:** {list or "none"}
-**Timestamp:** {ISO timestamp}
-```
+When `{proof_gate}=PASS`:
 
 ```bash
-bash {skill_dir}/scripts/update-progress.sh "{task_id}" "10" "verify" "complete"
+python3 "{skill_dir}/scripts/apex-state.py" event --root "$PWD" --run-id "{run_id}" --phase verify --status complete --message "Runtime proof gate passed"
+python3 "{skill_dir}/scripts/apex-state.py" checkpoint --root "$PWD" --run-id "{run_id}" --phase verify --message "Current proof artifacts recorded"
 ```
 
----
-
-## SUCCESS METRICS:
-
-✅ Recommended verification rules followed
-✅ Feature tested through real user surface
-✅ Each AC verified with actual interaction
-✅ 1+ screenshots captured as visual proof
-✅ Failures identified with clear descriptions
-✅ Fixes applied and re-verified (if needed)
-✅ User informed of verification status
-
-## FAILURE MODES:
-
-❌ Claiming verification passed without actually testing
-❌ Ignoring the project's recommended verification rules
-❌ No screenshot / visual proof of the feature working
-❌ Not launching the app
-❌ Testing only happy path, ignoring ACs
-❌ Auto-proceeding with FAIL verdict (without user OK)
-❌ Not comparing against original user request
-❌ **CRITICAL**: Not using AskUserQuestion for failure decisions
-
-## VERIFICATION PROTOCOLS:
-
-- Follow the project's recommended verification rules first (`.agents/rules/`, `AGENTS.md`, `CLAUDE.md`)
-- Test through REAL user surface (browser, CLI, API)
-- Capture 1+ screenshots as proof - no PASS verdict without evidence
-- Compare against ORIGINAL request, not just code
-- Fix and re-verify, don't just fix and hope
-- Be honest about failures
-
----
-
-## NEXT STEP:
-
-Based on flags:
-- **If pr_mode:** Load `./step-09-finish.md` to create pull request
-- **Otherwise:** Workflow complete - show summary
-
-<critical>
-Remember: The whole point of verify is to catch things that code review and tests miss.
-Test the feature like a real user would - don't just check the code!
-If teams_mode is active: NEVER shutdown teammates - they stay alive until step-09-finish!
-</critical>
+Then load `step-09-finish.md`.

package/agents-config/skills/appstore-connect/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Appstore Connect"
+  short_description: "Interact with App Store Connect via the asc CLI - apps,..."
+  icon_small: "./assets/codex-icon.svg"
+  icon_large: "./assets/codex-icon.svg"
+  brand_color: "#01C1C6"
+  default_prompt: "Use $appstore-connect to help with this task."

package/agents-config/skills/appstore-connect/assets/codex-icon.svg

@@ -0,0 +1,17 @@
+<!-- @license lucide-static v1.24.0 - ISC -->
+<svg role="img" aria-label="appstore-connect skill icon"
+  class="lucide lucide-cloud-upload"
+  xmlns="http://www.w3.org/2000/svg"
+  width="128"
+  height="128"
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="#F5F5F5"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+>
+  <path d="M12 13v8" />
+  <path d="M4 14.899A7 7 0 1 1 15.71 8h1.79a4.5 4.5 0 0 1 2.5 8.242" />
+  <path d="m8 17 4-4 4 4" />
+</svg>

package/agents-config/skills/commit/agents/openai.yaml

@@ -0,0 +1,10 @@
+interface:
+  display_name: "Commit"
+  short_description: "Quick commit and push with minimal, clean messages"
+  icon_small: "./assets/codex-icon.svg"
+  icon_large: "./assets/codex-icon.svg"
+  brand_color: "#4BD0FC"
+  default_prompt: "Use $commit to help with this task."
+
+policy:
+  allow_implicit_invocation: false

package/agents-config/skills/commit/assets/codex-icon.svg

@@ -0,0 +1,17 @@
+<!-- @license lucide-static v1.24.0 - ISC -->
+<svg role="img" aria-label="commit skill icon"
+  class="lucide lucide-git-commit-horizontal"
+  xmlns="http://www.w3.org/2000/svg"
+  width="128"
+  height="128"
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="#F5F5F5"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+>
+  <circle cx="12" cy="12" r="3" />
+  <line x1="3" x2="9" y1="12" y2="12" />
+  <line x1="15" x2="21" y1="12" y2="12" />
+</svg>

package/agents-config/skills/create-pr/agents/openai.yaml

@@ -0,0 +1,10 @@
+interface:
+  display_name: "Create Pr"
+  short_description: "Create and push PR with auto-generated title and description"
+  icon_small: "./assets/codex-icon.svg"
+  icon_large: "./assets/codex-icon.svg"
+  brand_color: "#951556"
+  default_prompt: "Use $create-pr to help with this task."
+
+policy:
+  allow_implicit_invocation: false

package/agents-config/skills/create-pr/assets/codex-icon.svg

@@ -0,0 +1,17 @@
+<!-- @license lucide-static v1.24.0 - ISC -->
+<svg role="img" aria-label="create-pr skill icon"
+  class="lucide lucide-git-merge"
+  xmlns="http://www.w3.org/2000/svg"
+  width="128"
+  height="128"
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="#F5F5F5"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+>
+  <circle cx="18" cy="18" r="3" />
+  <circle cx="6" cy="6" r="3" />
+  <path d="M6 21V9a9 9 0 0 0 9 9" />
+</svg>

package/agents-config/skills/environments-manager/SKILL.md

@@ -108,7 +108,7 @@ Base template, before backend-specific additions:
 set -euo pipefail
 
 WORKTREE_PATH="${CODEX_WORKTREE_PATH:-${CURSOR_WORKTREE_PATH:-$(pwd)}}"
-SOURCE_PATH="${CODEX_SOURCE_TREE_PATH:-${ROOT_WORKTREE_PATH:-$HOME/Developer/saas/<repo>}}"
+SOURCE_PATH="${CODEX_SOURCE_TREE_PATH:-${ROOT_WORKTREE_PATH:-$HOME/Developer/projects/<repo>}}"
 
 cd "$WORKTREE_PATH"
 

package/agents-config/skills/environments-manager/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Environments Manager"
+  short_description: "Set up per-worktree environments for Claude Code, Cursor, or..."
+  icon_small: "./assets/codex-icon.svg"
+  icon_large: "./assets/codex-icon.svg"
+  brand_color: "#444449"
+  default_prompt: "Use $environments-manager to help with this task."

package/agents-config/skills/environments-manager/assets/codex-icon.svg

@@ -0,0 +1,16 @@
+<!-- @license lucide-static v1.24.0 - ISC -->
+<svg role="img" aria-label="environments-manager skill icon"
+  class="lucide lucide-settings"
+  xmlns="http://www.w3.org/2000/svg"
+  width="128"
+  height="128"
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="#F5F5F5"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+>
+  <path d="M9.671 4.136a2.34 2.34 0 0 1 4.659 0 2.34 2.34 0 0 0 3.319 1.915 2.34 2.34 0 0 1 2.33 4.033 2.34 2.34 0 0 0 0 3.831 2.34 2.34 0 0 1-2.33 4.033 2.34 2.34 0 0 0-3.319 1.915 2.34 2.34 0 0 1-4.659 0 2.34 2.34 0 0 0-3.32-1.915 2.34 2.34 0 0 1-2.33-4.033 2.34 2.34 0 0 0 0-3.831A2.34 2.34 0 0 1 6.35 6.051a2.34 2.34 0 0 0 3.319-1.915" />
+  <circle cx="12" cy="12" r="3" />
+</svg>

package/agents-config/skills/environments-manager/examples/scripts/claude-worktree-remove.sh

@@ -13,7 +13,7 @@ CLEANUP_TIMEOUT_SEC="${CLAUDE_WORKTREE_CLEANUP_TIMEOUT:-300}"
 
 [[ ! -d "$WORKTREE" ]] && exit 0
 
-LOG_FILE="$WORKTREE/.worktree-cleanup.log"
+LOG_FILE="${TMPDIR:-/tmp}/claude-worktree-cleanup-$(basename "$WORKTREE").log"
 : > "$LOG_FILE" 2>/dev/null || LOG_FILE=""
 
 log_to() {
@@ -23,6 +23,18 @@ log_to() {
   fi
 }
 
+assert_clean_worktree() {
+  local changes
+  changes=$(git -C "$WORKTREE" status --porcelain --untracked-files=all 2>/dev/null || true)
+  if [[ -n "$changes" ]]; then
+    log_to "[claude-worktree-remove] REFUSED: worktree has uncommitted changes"
+    printf '%s\n' "$changes" >&2
+    exit 1
+  fi
+}
+
+assert_clean_worktree
+
 # Cleanup may call CLIs (e.g. convex env remove) that need the project's Node version.
 if [[ -s "$HOME/.nvm/nvm.sh" ]]; then
   # shellcheck disable=SC1090,SC1091
@@ -61,6 +73,10 @@ if [[ -x "${REPO}/scripts/worktree-down.sh" ]]; then
   wait "$WATCHDOG_PID" 2>/dev/null || true
 fi
 
+assert_clean_worktree
 BRANCH=$(git -C "$WORKTREE" rev-parse --abbrev-ref HEAD 2>/dev/null || true)
-git -C "$REPO" worktree remove "$WORKTREE" --force 2>/dev/null || true
-[[ -n "$BRANCH" && "$BRANCH" != "HEAD" ]] && git -C "$REPO" branch -D "$BRANCH" 2>/dev/null || true
+git -C "$REPO" worktree remove "$WORKTREE"
+if [[ -n "$BRANCH" && "$BRANCH" != "HEAD" ]]; then
+  git -C "$REPO" branch -d "$BRANCH" 2>/dev/null || \
+    log_to "[claude-worktree-remove] Kept unmerged branch: $BRANCH"
+fi

package/agents-config/skills/environments-manager/examples/scripts/worktree-up.sh

@@ -5,7 +5,7 @@
 set -euo pipefail
 
 WORKTREE_PATH="${CODEX_WORKTREE_PATH:-${CURSOR_WORKTREE_PATH:-$(pwd)}}"
-SOURCE_PATH="${CODEX_SOURCE_TREE_PATH:-${ROOT_WORKTREE_PATH:-$HOME/Developer/saas/myapp}}"
+SOURCE_PATH="${CODEX_SOURCE_TREE_PATH:-${ROOT_WORKTREE_PATH:-$HOME/Developer/projects/myapp}}"
 
 cd "$WORKTREE_PATH"
 

package/agents-config/skills/environments-manager/references/claude.md

@@ -99,7 +99,7 @@ These wrappers exist because Claude's `WorktreeCreate` hook has Claude-specific
 5. Run `scripts/worktree-up.sh` with `ROOT_WORKTREE_PATH`, `CLAUDE_WORKTREE_NAME`, and `CI=1` set, under a watchdog that hard-kills after `CLAUDE_WORKTREE_SETUP_TIMEOUT` (default 900s).
 6. Echo the worktree path on stdout — and absolutely nothing else.
 
-The remove wrapper mirrors steps 3–5 for cleanup, then runs `git worktree remove --force` + `git branch -D`.
+The remove wrapper mirrors steps 3–5 for cleanup, refuses a dirty worktree, then runs `git worktree remove` without force. It uses `git branch -d` and keeps any unmerged branch instead of destroying it.
 
 `ROOT_WORKTREE_PATH` is set so the shared `scripts/worktree-up.sh` can locate the source checkout via the existing cascade (`CODEX_SOURCE_TREE_PATH` → `ROOT_WORKTREE_PATH`).
 

package/agents-config/skills/fix-pr-comments/agents/openai.yaml

@@ -0,0 +1,10 @@
+interface:
+  display_name: "Fix Pr Comments"
+  short_description: "Fetch PR review comments and implement all requested changes"
+  icon_small: "./assets/codex-icon.svg"
+  icon_large: "./assets/codex-icon.svg"
+  brand_color: "#C62EE2"
+  default_prompt: "Use $fix-pr-comments to help with this task."
+
+policy:
+  allow_implicit_invocation: false

package/agents-config/skills/fix-pr-comments/assets/codex-icon.svg

@@ -0,0 +1,17 @@
+<!-- @license lucide-static v1.24.0 - ISC -->
+<svg role="img" aria-label="fix-pr-comments skill icon"
+  class="lucide lucide-git-merge"
+  xmlns="http://www.w3.org/2000/svg"
+  width="128"
+  height="128"
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="#F5F5F5"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+>
+  <circle cx="18" cy="18" r="3" />
+  <circle cx="6" cy="6" r="3" />
+  <path d="M6 21V9a9 9 0 0 0 9 9" />
+</svg>

package/agents-config/skills/grill-me/SKILL.md

@@ -1,10 +1,31 @@
 ---
 name: grill-me
-description: Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
+description: Grill the user in rapid batches to sharpen a plan, decision, or idea. Use when the user asks for Grill Me, wants a rigorous interview, or uses a grill trigger phrase.
+disable-model-invocation: true
 ---
 
-Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
+# Grill Me
 
-Ask the questions one at a time.
+Stress-test the user's thinking until both sides share a precise, actionable understanding.
 
-If a question can be answered by exploring the codebase, explore the codebase instead.
+## Interview loop
+
+1. Inspect the available environment first. Resolve discoverable facts through files, tools, documentation, or other in-scope evidence instead of asking the user.
+2. Open the session by recommending voice input: tell the user that answering with the microphone is fastest, and that short answers keyed `1` through `10` are enough.
+3. Ask exactly 10 numbered questions in one batch. Prioritize the highest-leverage unresolved decisions and order them so earlier questions clarify later ones.
+4. For every question, include a concise **Recommended answer** based on current evidence. Make the tradeoff or consequence clear enough for the user to accept, reject, or amend it quickly.
+5. Invite one grouped reply covering `1` through `10`. Accept terse answers, corrections, skipped items, or a blanket acceptance of the recommendations.
+6. After the reply, summarize what is established, call out contradictions or missing dependencies, and investigate any newly discoverable facts.
+7. Ask the next batch of 10 questions. Continue until the important branches of the decision tree are resolved. If fewer than 10 meaningful decisions remain, ask only those remaining; never add filler questions to reach 10.
+8. Present the final shared understanding as a compact decision brief: goal, scope, constraints, chosen approach, rejected alternatives, risks, and acceptance criteria.
+9. Ask for explicit confirmation that the brief is correct before acting on it.
+
+## Question quality
+
+- Ask decisions only the user can make; investigate facts yourself.
+- Challenge assumptions and compare credible alternatives instead of merely collecting preferences.
+- Keep each question independently answerable and concise enough for a spoken response.
+- Do not repeat settled questions unless new evidence invalidates the earlier answer.
+- Match the user's language.
+
+Do not implement the plan until the user confirms the final decision brief.

package/agents-config/skills/grill-me/agents/openai.yaml

@@ -0,0 +1,8 @@
+interface:
+  display_name: "Grill Me"
+  short_description: "Stress-test ideas in rapid 10-question rounds"
+  icon_small: "./assets/codex-icon.svg"
+  icon_large: "./assets/codex-icon.svg"
+  brand_color: "#F5F5F5"
+policy:
+  allow_implicit_invocation: false

package/agents-config/skills/grill-me/assets/codex-icon.svg

@@ -0,0 +1,16 @@
+<!-- @license lucide-static v0.294.0 - ISC -->
+<svg role="img" aria-label="grill-me skill icon"
+  class="lucide lucide-messages-square"
+  xmlns="http://www.w3.org/2000/svg"
+  width="128"
+  height="128"
+  viewBox="0 0 24 24"
+  fill="none"
+  stroke="#F5F5F5"
+  stroke-width="2"
+  stroke-linecap="round"
+  stroke-linejoin="round"
+>
+  <path d="M14 9a2 2 0 0 1-2 2H6l-4 4V4c0-1.1.9-2 2-2h8a2 2 0 0 1 2 2v5Z" />
+  <path d="M18 9h2a2 2 0 0 1 2 2v11l-4-4h-6a2 2 0 0 1-2-2v-1" />
+</svg>