fix(cli): replace stale full-auto guidance

Author: BaseInfinity

README.md

@@ -10,15 +10,15 @@ This adapter brings the [SDLC Wizard](https://github.com/BaseInfinity/agentic-ai
 # Setup a new repo or sync an already-initialized clone
 npx codex-sdlc-wizard@latest
 
-# Start coding with SDLC enforcement
-codex --full-auto
+# Start coding with SDLC enforcement and an explicit model profile
+codex -m gpt-5.5 -c 'model_reasoning_effort="xhigh"'
 ```
 
-`codex --full-auto` is the recommended default once this wizard is installed: you keep the repo guardrails and hook enforcement, but day-to-day editing and runs stay low-friction. Use plain `codex` instead if you want more manual confirmation. If a handoff is interrupted and Codex prints a resume id, continue with `codex resume --full-auto <session-id>` for the same low-friction posture.
+`codex -m gpt-5.5 -c 'model_reasoning_effort="xhigh"'` is the safest explicit start once this wizard is installed. Use plain `codex` instead if you want to rely on trusted repo-local config. If a handoff is interrupted and Codex prints a resume id, continue with `codex resume -m gpt-5.5 -c 'model_reasoning_effort="xhigh"' <session-id>` so resume does not fall back to an older model.
 
-If you normally use yolo-style sessions, use Codex's explicit current flags instead: `--sandbox danger-full-access --ask-for-approval never`. Only use that full-trust variant in repos you fully trust.
+If you normally use yolo-style sessions, use Codex's canonical full-trust flag: `--dangerously-bypass-approvals-and-sandbox`. Current Codex may accept `--yolo` as shorthand, but this wizard prints the canonical flag. Full-auto is not full-trust: full-trust bypasses sandbox and approval prompts. Only use that variant in repos you fully trust.
 
-Bare `npx codex-sdlc-wizard` is the adaptive setup/sync path. In an already-initialized repo clone, it runs the update/check-repair path automatically so a fresh Mac/Windows/Linux checkout can sync hooks, config, and helper skills without remembering separate commands. In a new repo, it bootstraps the repo-local guardrails first, then hands off into a live plain Codex setup session so the unresolved setup questions happen inside Codex instead of inside a shell checklist. At that first-run handoff prompt, press Enter for plain `codex` or type `full-auto` if you explicitly want `codex --full-auto`. `setup --yes` still exists for automation, but it is not the normal human path.
+Bare `npx codex-sdlc-wizard` is the adaptive setup/sync path. In an already-initialized repo clone, it runs the update/check-repair path automatically so a fresh Mac/Windows/Linux checkout can sync hooks, config, and helper skills without remembering separate commands. In a new repo, it bootstraps the repo-local guardrails first, then hands off into a live plain Codex setup session so the unresolved setup questions happen inside Codex instead of inside a shell checklist. At that first-run handoff prompt, press Enter for plain `codex` or type `full-trust` if you explicitly want `codex --dangerously-bypass-approvals-and-sandbox`. `setup --yes` still exists for automation, but it is not the normal human path.
 
 Generic npm entrypoint examples: `npx codex-sdlc-wizard`, `npx codex-sdlc-wizard check`, and `npx codex-sdlc-wizard update`.
 

ROADMAP.md

@@ -19,10 +19,10 @@
 - setup/update guidance now treats verification as diagnostic for product failures and stops before editing application code or application tests without explicit user consent
 - setup/update guidance now tells users to exit and reopen Codex after hook/skill repairs, without rerunning setup/update just for that restart
 - install/setup/update now write and repair repo-local `.codex/config.toml` model keys for the selected profile, while preserving unrelated MCP, sandbox, approval, and custom config
-- first-run live setup now defaults to plain `codex` after bootstrap and requires an explicit `full-auto` choice to start that setup handoff with `codex --full-auto`
-- first-run handoff now uses a clearer prompt, recommends `codex resume --full-auto` for interrupted handoffs, and avoids the deprecated Windows `shell:true` plus args launcher path
+- first-run live setup now defaults to plain `codex` after bootstrap and requires an explicit `full-trust` choice to start that setup handoff with `codex --dangerously-bypass-approvals-and-sandbox`
+- first-run handoff now uses a clearer prompt, recommends model-explicit `codex resume -m ... -c ...` for interrupted handoffs, and avoids the deprecated Windows `shell:true` plus args launcher path
 - first-run live Codex handoff now runs as a managed child process with opt-in timeout cleanup, POSIX signal forwarding, process-group termination, repeated-interrupt handling, and explicit retry/resume guidance
-- setup/install output now prints explicit full-trust Codex flags (`--sandbox danger-full-access --ask-for-approval never`) for users who normally say yolo-style sessions
+- setup/install output now prints Codex's canonical full-trust flag (`--dangerously-bypass-approvals-and-sandbox`) for users who normally say yolo-style sessions, while keeping full-trust distinct from historical full-auto wording
 - update guidance now frontloads the npm version boundary: `$update-wizard` repairs repo artifacts, while `npx codex-sdlc-wizard@latest update` consumes the newest package
 - setup guidance now includes Codex Desktop handoff notes for auth-heavy browser/computer-use setup flows
 - generated setup docs and shipped skills now include a task-routing gate that identifies CLI, Desktop/computer-use, browser automation, or human-only lanes before giving execution steps

SKILL.md

@@ -17,7 +17,7 @@ Use the bundled scripts like this:
 1. If the user wants the simplest install, run the bundled `install.sh` from this skill bundle against the current working repo.
 2. If the user wants repo-aware setup or regenerated docs, run the bundled `setup.sh --yes` from this skill bundle against the current working repo.
 3. Tell the user exactly which path you chose: `install.sh` for baseline enforcement, `setup.sh` for adaptive setup.
-4. After installation, tell the user to start a fresh Codex session so hooks and repo docs are loaded cleanly. Recommend plain `codex` for the first live setup handoff unless the user explicitly chooses `full-auto`; recommend `codex --full-auto` as the default start mode after setup is complete and the guardrails are loaded. If a handoff was interrupted and Codex printed a resume id, recommend `codex resume --full-auto <session-id>` for low-friction continuation. Mention plain `codex` or `codex resume <session-id>` as the manual fallback for daily work.
+4. After installation, tell the user to start a fresh Codex session so hooks and repo docs are loaded cleanly. Recommend plain `codex` for the first live setup handoff unless the user explicitly chooses `full-trust`; recommend `codex -m <model> -c 'model_reasoning_effort="xhigh"'` as the current explicit start mode after setup is complete and the guardrails are loaded. If a handoff was interrupted and Codex printed a resume id, recommend `codex resume -m <model> -c 'model_reasoning_effort="xhigh"' <session-id>` for model-explicit continuation. Mention plain `codex` or `codex resume <session-id>` as the config-driven fallback for daily work. For yolo-style sessions, use the canonical full-trust flag `--dangerously-bypass-approvals-and-sandbox`; current Codex may accept `--yolo` as shorthand, but full-auto is not full-trust.
    The fresh session should also pick up the repo-scoped `\$sdlc` skill under `.agents/skills`. Repo-scoped skill coverage is still a work in progress; additional workflow names should stay out of public handoff copy until their contracts are ready.
    Canonical SDLC entrypoint: `\$sdlc`. `/sdlc` is historical shorthand for the missing slash-command idea, not an invocation command.
    For setup/update bootstrap work, recommend the `maximum` profile via `--model-profile maximum` as the safer default. For routine work after bootstrap, point users back to the `mixed` profile via `--model-profile mixed` for the better speed / lower latency / lower token path with `xhigh` review.

bin/codex-sdlc-wizard.js

@@ -14,6 +14,7 @@ const interactiveSessionPrompt = [
   "Continue setup inside Codex: scan the repo, ask only unresolved questions, preserve intentional existing docs, generate or refresh repo-specific SDLC docs, verify the result, and finish setup.",
   "Use xhigh reasoning for setup."
 ].join(" ");
+const fullTrustFlag = "--dangerously-bypass-approvals-and-sandbox";
 
 function printHelp() {
   process.stdout.write(`Usage: codex-sdlc-wizard [setup|check|update|install] [options]
@@ -25,7 +26,8 @@ Commands:
   install        Advanced escape hatch: run install.sh without adaptive setup
 
 Default behavior: initialized repos run update automatically; new repos use adaptive setup, then hand off into a live plain Codex setup session.
-Type "full-auto" at the handoff prompt if you want codex --full-auto for first-run setup.
+Type "full-trust" at the handoff prompt if you want codex ${fullTrustFlag} for first-run setup.
+Full-trust/yolo is separate from automation posture; it bypasses sandbox and approval prompts.
 Automation/non-interactive behavior: use setup --yes to stay on the shell path.
 Bootstrap/setup recommendation: maximum.
 Routine work after bootstrap: mixed.
@@ -207,7 +209,8 @@ function printHandoffRecovery(reason) {
     reason,
     "Terminating spawned Codex process tree.",
     "Retry from this repo with: npx codex-sdlc-wizard@latest",
-    "If Codex printed a session id before stopping, resume with: codex resume --full-auto -m gpt-5.5 -c 'model_reasoning_effort=\"xhigh\"' <session-id>",
+    "If Codex printed a session id before stopping, resume with: codex resume -m gpt-5.5 -c 'model_reasoning_effort=\"xhigh\"' <session-id>",
+    `For full-trust/yolo-style resume, use: codex resume ${fullTrustFlag} -m gpt-5.5 -c 'model_reasoning_effort=\"xhigh\"' <session-id>`,
     ""
   ].join("\n"));
 }
@@ -404,8 +407,8 @@ function runScript(scriptName, args) {
 }
 
 async function askHandoffMode() {
-  if (process.env.CODEX_SDLC_HANDOFF_MODE === "full-auto") {
-    return "full-auto";
+  if (process.env.CODEX_SDLC_HANDOFF_MODE === "full-trust" || process.env.CODEX_SDLC_HANDOFF_MODE === "yolo") {
+    return "full-trust";
   }
 
   if (process.env.CODEX_SDLC_HANDOFF_MODE === "plain") {
@@ -416,16 +419,17 @@ async function askHandoffMode() {
     "",
     "Choose first-run Codex handoff mode:",
     "  Press Enter: plain codex (recommended)",
-    '  Type "full-auto": codex --full-auto',
-    "  If interrupted later, resume with: codex resume --full-auto <session-id>",
+    `  Type "full-trust": codex ${fullTrustFlag}`,
+    "  If you say yolo, use full-trust; full-auto is not full-trust.",
+    "  If interrupted later, resume with: codex resume -m gpt-5.5 -c 'model_reasoning_effort=\"xhigh\"' <session-id>",
     "> "
   ].join("\n");
 
   if (!process.stdin.isTTY) {
     process.stdout.write(prompt);
     const answer = fs.readFileSync(0, "utf8").split(/\r?\n/, 1)[0].trim().toLowerCase();
     process.stdout.write("\n");
-    return answer === "full-auto" ? "full-auto" : "plain";
+    return answer === "full-trust" || answer === "yolo" ? "full-trust" : "plain";
   }
 
   const rl = readline.createInterface({
@@ -435,7 +439,7 @@ async function askHandoffMode() {
 
   try {
     const answer = (await rl.question(prompt)).trim().toLowerCase();
-    return answer === "full-auto" ? "full-auto" : "plain";
+    return answer === "full-trust" || answer === "yolo" ? "full-trust" : "plain";
   } finally {
     rl.close();
   }
@@ -455,7 +459,7 @@ async function handoffToCodex(modelProfile) {
   }
 
   const handoffMode = await askHandoffMode();
-  const modeLabel = handoffMode === "full-auto" ? "codex --full-auto" : "plain codex";
+  const modeLabel = handoffMode === "full-trust" ? `codex ${fullTrustFlag}` : "plain codex";
   process.stdout.write(`\nHanding off into Codex for live setup using ${modeLabel}...\n`);
 
   const codexArgs = [
@@ -468,8 +472,8 @@ async function handoffToCodex(modelProfile) {
     interactiveSessionPrompt
   ];
 
-  if (handoffMode === "full-auto") {
-    codexArgs.unshift("--full-auto");
+  if (handoffMode === "full-trust") {
+    codexArgs.unshift(fullTrustFlag);
   }
 
   const codexResult = await runCodexHandoff(codexArgs);

install.ps1

@@ -243,17 +243,19 @@ Write-Host ""
 Write-Host "SDLC Wizard for Codex installed."
 $startModel = if ($ModelProfile -eq "maximum") { "gpt-5.5" } else { "gpt-5.4-mini" }
 $startReasoning = "xhigh"
-Write-Host "Recommended start: 'codex --full-auto' for low-friction SDLC inside the repo guardrails."
-Write-Host "Use plain 'codex' instead if you want more manual confirmation."
+Write-Host "Recommended start: codex -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
+Write-Host "Use plain 'codex' instead if you want to rely on trusted repo-local config."
 Write-Host "Fresh-session note: if you ran this from inside an existing Codex session, exit and reopen Codex in this repo so repo-local config, hooks, and skills load."
 Write-Host "Hook review note: if Codex says hooks need review, open /hooks after restart and review pending repo hooks before relying on enforcement."
-Write-Host "Start new with selected profile: codex --full-auto -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
-Write-Host "Resume with selected profile: codex resume --full-auto -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
-Write-Host "If resume warns it came back with a different model, resume explicitly with: codex resume --full-auto -m gpt-5.5 -c 'model_reasoning_effort=`"xhigh`"'"
-Write-Host "If you normally use yolo-style sessions, use the explicit current Codex flags:"
-Write-Host "  codex -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"' --sandbox danger-full-access --ask-for-approval never"
-Write-Host "  codex resume -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"' --sandbox danger-full-access --ask-for-approval never"
-Write-Host "Full-trust warning: only use that variant in repos you fully trust; it bypasses sandbox and approval prompts."
+Write-Host "Start new with selected profile: codex -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
+Write-Host "Resume with selected profile: codex resume -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
+Write-Host "If resume warns it came back with a different model, resume explicitly with: codex resume -m gpt-5.5 -c 'model_reasoning_effort=`"xhigh`"'"
+Write-Host "If you normally use yolo-style sessions, use the canonical full-trust Codex flag:"
+Write-Host "  codex --dangerously-bypass-approvals-and-sandbox -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
+Write-Host "  codex resume --dangerously-bypass-approvals-and-sandbox -m $startModel -c 'model_reasoning_effort=`"$startReasoning`"'"
+Write-Host "Codex may accept --yolo as shorthand; this wizard prints the canonical full-trust flag."
+Write-Host "Full-auto is not full-trust: full-trust bypasses sandbox and approval prompts."
+Write-Host "Full-trust warning: only use that variant in repos you fully trust."
 Write-Host "Recommended: use full access during setup, environment repair, and auth-heavy workflows."
 Write-Host "Wrote repo-local .codex/config.toml model keys for this profile; mixed is wizard policy, not a native Codex mode."
 Write-Host "Codex loads project config only after the repo is trusted, and trusted project config overrides your user-level ~/.codex/config.toml."

install.sh

@@ -256,19 +256,21 @@ write_model_profile
 
 echo ""
 echo "SDLC Wizard for Codex installed."
-echo "Recommended start: 'codex --full-auto' for low-friction SDLC inside the repo guardrails."
-echo "Use plain 'codex' instead if you want more manual confirmation."
 START_MODEL="$(profile_model "$MODEL_PROFILE")"
 START_REASONING="$(profile_reasoning "$MODEL_PROFILE")"
+echo "Recommended start: codex -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
+echo "Use plain 'codex' instead if you want to rely on trusted repo-local config."
 echo "Fresh-session note: if you ran this from inside an existing Codex session, exit and reopen Codex in this repo so repo-local config, hooks, and skills load."
 echo "Hook review note: if Codex says hooks need review, open /hooks after restart and review pending repo hooks before relying on enforcement."
-echo "Start new with selected profile: codex --full-auto -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
-echo "Resume with selected profile: codex resume --full-auto -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
-echo "If resume warns it came back with a different model, resume explicitly with: codex resume --full-auto -m gpt-5.5 -c 'model_reasoning_effort=\"xhigh\"'"
-echo "If you normally use yolo-style sessions, use the explicit current Codex flags:"
-echo "  codex -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"' --sandbox danger-full-access --ask-for-approval never"
-echo "  codex resume -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"' --sandbox danger-full-access --ask-for-approval never"
-echo "Full-trust warning: only use that variant in repos you fully trust; it bypasses sandbox and approval prompts."
+echo "Start new with selected profile: codex -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
+echo "Resume with selected profile: codex resume -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
+echo "If resume warns it came back with a different model, resume explicitly with: codex resume -m gpt-5.5 -c 'model_reasoning_effort=\"xhigh\"'"
+echo "If you normally use yolo-style sessions, use the canonical full-trust Codex flag:"
+echo "  codex --dangerously-bypass-approvals-and-sandbox -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
+echo "  codex resume --dangerously-bypass-approvals-and-sandbox -m $START_MODEL -c 'model_reasoning_effort=\"$START_REASONING\"'"
+echo "Codex may accept --yolo as shorthand; this wizard prints the canonical full-trust flag."
+echo "Full-auto is not full-trust: full-trust bypasses sandbox and approval prompts."
+echo "Full-trust warning: only use that variant in repos you fully trust."
 echo "Model profile: '$MODEL_PROFILE'."
 echo "  - mixed: gpt-5.4-mini main pass + gpt-5.5 xhigh review for better speed, lower latency, and lower token usage."
 echo "  - maximum: gpt-5.5 xhigh throughout for maximum stability and the most thorough \"ultimate mode\"."

skills/setup-wizard/SKILL.md

@@ -183,7 +183,7 @@ This verification is diagnostic for product behavior. If a failing command point
 
 ### Step 8: Restart and next steps
 
-If new skills or hooks were installed or repaired, tell the user to exit and reopen Codex in this repo so the active session reloads them. Tell them: you do not need to rerun setup just for that restart. If they closed an interrupted handoff and Codex printed a resume id, recommend `codex resume --full-auto <session-id>` for low-friction continuation, or plain `codex resume <session-id>` when they want manual confirmations.
+If new skills or hooks were installed or repaired, tell the user to exit and reopen Codex in this repo so the active session reloads them. Tell them: you do not need to rerun setup just for that restart. If they closed an interrupted handoff and Codex printed a resume id, recommend `codex resume -m <model> -c 'model_reasoning_effort="xhigh"' <session-id>` for model-explicit continuation, or plain `codex resume <session-id>` when they want config-driven startup. If they normally say yolo, give the full-trust variant with `--dangerously-bypass-approvals-and-sandbox` and say full-auto is not full-trust.
 
 Then point them at the next entrypoint: