diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000..c9f4405
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,55 @@
+# Changelog
+
+## 2.1.1 — 2026-07-15
+
+Robustness release: script hardening, dynamic branching, and a regression test suite.
+
+### Fixed
+
+- **State-directory footgun**: `codex-implement`, `codex-code-review`, and `codex-ask` now ship their own `start`/`resume`/`reset`/`show` wrapper scripts that pin `STATE_DIR` themselves. The old `export STATE_DIR` choreography silently broke when the export and the script call ran in separate shells — worst case, a resume continued the *plan-review* thread with the reviewer model instead of the implementation thread.
+- **State-key lookup for free-form labels**: `TRIP-3-release` now derives the code-review state key via the new shared `key.sh` instead of an inline `realpath` pipeline, which mis-keyed unplanned work (free-form labels) and silently downgraded a converged Codex review to the "skipped" fallback.
+- **Prompt substitution**: `load_prompt` no longer corrupts prompts containing `&` or backslashes, and no longer loops forever when a substituted value quotes its own placeholder (e.g. implementer notes discussing the templates).
+- **Committed session artifacts**: added the missing `state/.gitignore` for `codex-implement` and `codex-ask`, so `git add -A` during release no longer commits thread IDs, reports, and event logs.
+- **Stale v1 references**: week-anchor formula location, synthesize template's version-step pointer, tutorial step numbering and its renumber list.
+
+### Changed
+
+- **Dynamic base branch**: `TRIP-2-implement`, `TRIP-3-release`, and `TRIP-hotfix` now work from whatever branch is active. The base branch is recorded at branch creation (`branch.<name>.trip-base` git config) and the release/hotfix merge returns to it — the `[MAIN_BRANCH]` placeholder is gone.
+- **Push is now explicit**: `TRIP-3-release` and `TRIP-hotfix` list the available branches and ask which one to push (base branch recommended), unless the user already specified one. Hotfix previously pushed automatically as part of the merge chain.
+- **Hotfix merge policy**: `TRIP-hotfix` now enforces `--ff-only` like the release skill — never a merge commit.
+- **Docs**: unified ARCHI.md token thresholds (10–15k target, ~20k ceiling) across README, `TRIP-compact`, and `TRIP-3-release`; added the Bash/Codex CLI/`jq`/GNU-coreutils requirements note; typo fixes.
+
+### Added
+
+- `tests/workflow-regression.sh` — regression suite covering prompt substitution (literal `&`/backslash handling, self-referencing placeholder termination), shared key derivation for free-form labels, wrapper state-directory isolation, and state `.gitignore` presence.
+
+## 2.1.0 — 2026-07-13
+
+### Added
+
+- **`codex-ask` skill**: grounded second opinion from Codex on any matter — architecture calls, debugging hypotheses, research conclusions. Read-only sandbox, threaded per topic, advisory only (no verdict tags, nothing gated).
+- **TRIP-research cross-check step**: decision-grade findings are red-teamed via `codex-ask` before being presented; real disagreements are recorded in the memo's Open Questions section.
+
+## 2.0.0 — 2026-07-13
+
+Flow simplified to **Plan → Implement → Release**: review and test moved inside Implement as a testing gate and an automatic Codex review loop.
+
+### Added
+
+- **`codex-implement` skill**: implementation delegated to Codex CLI in a workspace-write sandbox, with persistent per-plan threads for phased delegation.
+- **`TRIP-3-release` skill**: owns the release ceremony (version, CR promotion, changelogs, docs, commit, tag, ff-merge, push).
+- **Testing gate** in `TRIP-2-implement`, with the hard-to-cover policy and coverage-debt ledger.
+- **Per-flow Codex model defaults** in `codex-plan-review/scripts/_common.sh` (implementation vs review models), overridable via `CODEX_MODEL` / `CODEX_EFFORT`.
+
+### Changed
+
+- `TRIP-review` and `TRIP-test` reborn as on-demand support skills (manual fallback/audit review, deep test-authoring reference).
+- `TRIP-upgrade` handles the v1 → v2 rename/migration (`TRIP-3-review` → `TRIP-review`, `TRIP-4-test` → `TRIP-test`).
+
+## Earlier (1.x)
+
+- 2026-06-09 — Codex review integration, `TRIP-upgrade` skill, review file separation.
+- 2026-02-10 — minor tool call update.
+- 2026-02-05 — Mistral Vibe compatibility, contributing section, `AskUserQuestion` emulation skill.
+- 2026-02-03 — migration to Skills.
+- 2026-01-29 — initial commit.
diff --git a/README.md b/README.md
index 716ee2c..f526164 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,6 @@
 ![TRIP Workflow Banner](assets/trip-workflow-banner2.png)
 
-![Version](https://img.shields.io/badge/version-2.1.0-blue) [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://github.com/PiLastDigit/TRIP-workflow/blob/master/LICENSE) ![Works with](https://img.shields.io/badge/Works_with-grey) [![Claude Code](https://img.shields.io/badge/Claude_Code-E5582B)](https://docs.anthropic.com/en/docs/claude-code) [![Codex CLI](https://img.shields.io/badge/Codex_CLI-10A37F)](https://developers.openai.com/codex/cli/) [![OpenCode](https://img.shields.io/badge/OpenCode-1a3a5c)](https://github.com/sst/opencode) [![Mistral Vibe](https://img.shields.io/badge/Mistral_Vibe-F7D046)](https://github.com/mistralai/mistral-vibe)
+![Version](https://img.shields.io/badge/version-2.1.1-blue) [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://github.com/PiLastDigit/TRIP-workflow/blob/master/LICENSE) ![Works with](https://img.shields.io/badge/Works_with-grey) [![Claude Code](https://img.shields.io/badge/Claude_Code-E5582B)](https://docs.anthropic.com/en/docs/claude-code) [![Codex CLI](https://img.shields.io/badge/Codex_CLI-10A37F)](https://developers.openai.com/codex/cli/) [![OpenCode](https://img.shields.io/badge/OpenCode-1a3a5c)](https://github.com/sst/opencode) [![Mistral Vibe](https://img.shields.io/badge/Mistral_Vibe-F7D046)](https://github.com/mistralai/mistral-vibe)
 
 ## What is TRIP?
 
@@ -18,7 +18,7 @@ Even the "simple" ones come with:
 
 - 47 different commands & skills to memorize
 - Sub-agents swarm for God-knows-what
-- Mutlti-chapters courses (sometimes paid lol)
+- Multi-chapter courses (sometimes paid lol)
 
 **TRIP is different.** It's deliberately minimal:
 
@@ -38,6 +38,8 @@ It was kept stupid simple because **the goal is to ship features, not to master
 
 ## Getting Started
 
+Requirements for the bundled Codex scripts: Bash, Codex CLI, `jq`, and GNU coreutils (`realpath` and GNU `date`). On Windows, run them from WSL or a shell with those tools installed.
+
 1. Copy the `skills/` folder contents to your repo's `.claude/skills/` or whatever
 2. Run `/TRIP-init [YourProjectName]`
 3. Follow the interactive prompts
@@ -132,7 +134,7 @@ Exploratory investigation with defined compute level. For feasibility studies an
 ### `/TRIP-compact`
 
 Run this skill to compact ARCHI.md size while preserving relevance, accuracy, and coverage through summarization and restructuring. Token calculator script included.
-As a rule of thumb, ARCHI.md should not exceed ~10% of context window.
+Aim for 10,000–15,000 tokens; treat ~20,000 as the hard ceiling.
 
 ## Multi-Agent: Using Different LLMs at Different Steps
 
@@ -147,7 +149,7 @@ As of mid july 2026, this Fable + GPT5.6 harness combo is absolute peak.
 
 ## MCP Servers: Less Is More
 
-Last piece of advise before your new coding quest: Every MCP server you add is extra context, extra latency, and extra confusion. Keep it minimal. The one use case where MCP genuinely shines is **up-to-date documentation**, so your agent stops hallucinating deprecated APIs/whatever. Two servers cover it: [Context7](https://github.com/upstash/context7) for current library & framework docs, and [Exa](https://github.com/exa-labs/exa-mcp-server) for web search when the answer isn't in any doc. No bloat beyond that.
+Last piece of advice before your new coding quest: Every MCP server you add is extra context, extra latency, and extra confusion. Keep it minimal. The one use case where MCP genuinely shines is **up-to-date documentation**, so your agent stops hallucinating deprecated APIs/whatever. Two servers cover it: [Context7](https://github.com/upstash/context7) for current library & framework docs, and [Exa](https://github.com/exa-labs/exa-mcp-server) for web search when the answer isn't in any doc. No bloat beyond that.
 
 ## Contributing
 
diff --git a/skills/TRIP-2-implement/SKILL.md b/skills/TRIP-2-implement/SKILL.md
index e1b72de..253ead4 100644
--- a/skills/TRIP-2-implement/SKILL.md
+++ b/skills/TRIP-2-implement/SKILL.md
@@ -22,13 +22,15 @@ Implement: $ARGUMENTS
 
 ## Step 0: Create a Branch (Pre-Implementation)
 
-**Always** create a dedicated branch before implementing — no need to ask. `TRIP-3-release` merges it back into the main branch with fast-forward, keeping a single clean linear history.
+**Always** create a dedicated branch before implementing — no need to ask. The workflow is base-branch agnostic: whatever branch is active right now becomes the **base branch**, and `TRIP-3-release` merges the feature branch back into it with fast-forward, keeping a single clean linear history.
 
 ```bash
+BASE_BRANCH="$(git branch --show-current)"
 git checkout -b feat/[short-description]   # or fix/[short-description]
+git config branch."feat/[short-description]".trip-base "$BASE_BRANCH"
 ```
 
-Derive the short description from the plan/feature name. If already on a dedicated branch for this work (e.g., resuming a session), continue on it.
+The `trip-base` config entry records where the branch came from so the release step knows where to merge back — no hardcoded main. Derive the short description from the plan/feature name. If already on a dedicated branch for this work (e.g., resuming a session), continue on it.
 
 ---
 
@@ -49,8 +51,7 @@ You do NOT write the implementation yourself — delegate it to Codex via the `c
    Follow-up phases resume the same thread (context retained):
 
    ```bash
-   export STATE_DIR=".claude/skills/codex-implement/state"
-   bash .claude/skills/codex-plan-review/scripts/resume.sh \
+   bash .claude/skills/codex-implement/scripts/resume.sh \
        --prompt-file .claude/skills/codex-implement/prompts/continue.tpl \
        <plan-path> "Now implement Phase 2"
    ```
@@ -126,15 +127,9 @@ Always run the Codex code review after the testing gate passes — no confirmati
 
 ### Loop
 
-Always export before invoking shared scripts:
-
-```bash
-export STATE_DIR=".claude/skills/codex-code-review/state"
-```
-
 1. **Start**:
    ```bash
-   bash .claude/skills/codex-plan-review/scripts/start.sh \
+   bash .claude/skills/codex-code-review/scripts/start.sh \
        --prompt-file .claude/skills/codex-code-review/prompts/start.tpl \
        <plan-path> "$GATE_SUMMARY"
    ```
@@ -148,7 +143,7 @@ export STATE_DIR=".claude/skills/codex-code-review/state"
 
 5. **Resume** (re-run the testing gate first — lint, typecheck, affected tests — and build a fresh summary):
    ```bash
-   bash .claude/skills/codex-plan-review/scripts/resume.sh \
+   bash .claude/skills/codex-code-review/scripts/resume.sh \
        --prompt-file .claude/skills/codex-code-review/prompts/resume.tpl \
        --notes "Fixed X. Pushed back on Y because Z." \
        <plan-path> "$GATE_SUMMARY"
@@ -164,7 +159,7 @@ Skip if loop converged on Turn 1 (state file already holds full review).
 Turn-N state files hold only that turn's delta. After multi-round convergence, produce a consolidated review:
 
 ```bash
-bash .claude/skills/codex-plan-review/scripts/resume.sh \
+bash .claude/skills/codex-code-review/scripts/resume.sh \
     --prompt-file .claude/skills/codex-code-review/prompts/synthesize.tpl \
     <plan-path> "Today's date is YYYY-MM-DD"
 ```
diff --git a/skills/TRIP-3-release/SKILL.md b/skills/TRIP-3-release/SKILL.md
index d854af0..50886dd 100644
--- a/skills/TRIP-3-release/SKILL.md
+++ b/skills/TRIP-3-release/SKILL.md
@@ -60,7 +60,7 @@ Now that week (`a`) and version (`x.y.z`) are known:
 
 1. Compute state file path:
    ```bash
-   STATE_KEY="$(realpath <plan-path> | sed 's|^/||; s|/|__|g')"
+   STATE_KEY="$(bash .claude/skills/codex-plan-review/scripts/key.sh <plan-path-or-label>)"
    STATE_FILE=".claude/skills/codex-code-review/state/${STATE_KEY}.review.txt"
    ```
 
@@ -158,25 +158,36 @@ git tag vx.y.z
 
 ## Step 11: Merge (fast-forward)
 
-Merge the feature branch back into the main branch, keeping a single clean linear history:
+Merge the feature branch back into its **base branch** — the branch it was created from, recorded by `TRIP-2-implement` Step 0 — keeping a single clean linear history. Run as ONE command so the lookup and the merge share a shell:
 
 ```bash
-git checkout [MAIN_BRANCH]
-git merge --ff-only <feature-branch>
-git branch -d <feature-branch>
+BASE_BRANCH="$(git config branch.<feature-branch>.trip-base)" \
+    && git checkout "$BASE_BRANCH" \
+    && git merge --ff-only <feature-branch> \
+    && git branch -d <feature-branch>
 ```
 
-If `--ff-only` fails, the main branch moved during implementation — rebase the feature branch onto it, then retry. **Never create a merge commit.**
+If the `trip-base` lookup fails (branch created outside TRIP-2), ask the user which branch to merge into and substitute it for `"$BASE_BRANCH"`.
+
+If `--ff-only` fails, the base branch moved during implementation — rebase the feature branch onto it, then retry. **Never create a merge commit.**
 
 ## Step 12: Push
 
-**Use the `AskUserQuestion` tool** to ask:
+**If the user already specified which branch to push** (in the request or earlier in the session), push that branch directly and skip the question.
+
+Otherwise, list the available branches:
+
+```bash
+git branch --format='%(refname:short)'
+```
+
+Then **use the `AskUserQuestion` tool** to ask:
 
-- **Question**: "Release vx.y.z is committed, tagged, and merged. Push to remote?"
-- **Options**: "Yes, push now" (push branch and tags), "Not yet" (push manually later)
+- **Question**: "Release vx.y.z is committed, tagged, and merged. Which branch should I push?"
+- **Options**: One option per branch from the list above — the base branch first, marked "(recommended)" — plus "Not yet" (push manually later). If there are too many branches to list, include the most relevant ones; the user can name another via "Other".
 
-**If "Yes"**:
+**If a branch is selected**:
 
 ```bash
-git push && git push --tags
+git push origin <selected-branch> && git push --tags
 ```
diff --git a/skills/TRIP-compact/SKILL.md b/skills/TRIP-compact/SKILL.md
index 9ee2727..6d59c98 100644
--- a/skills/TRIP-compact/SKILL.md
+++ b/skills/TRIP-compact/SKILL.md
@@ -10,7 +10,7 @@ You are now in **compaction mode** - intelligently reducing ARCHI.md size while
 
 ## Why Compact?
 
-ARCHI.md should not exceed _~20k tokens_. A bloated ARCHI:
+Aim for 10,000–15,000 tokens and treat ~20,000 as the hard ceiling. A bloated ARCHI:
 
 - Consumes tokens that could be used for actual work
 - Slows down every command that reads it
diff --git a/skills/TRIP-compact/count-tokens.sh b/skills/TRIP-compact/count-tokens.sh
old mode 100755
new mode 100644
diff --git a/skills/TRIP-hotfix/SKILL.md b/skills/TRIP-hotfix/SKILL.md
index 617086a..36727b8 100644
--- a/skills/TRIP-hotfix/SKILL.md
+++ b/skills/TRIP-hotfix/SKILL.md
@@ -34,11 +34,17 @@ Before proceeding, confirm this is a genuine hotfix:
 
 ## Step 2: Create Hotfix Branch
 
+Branch from the **currently active branch** — the workflow is base-branch agnostic, and Step 8 merges back into whatever branch you started from. Run as ONE command so the steps share a shell:
+
 ```bash
-git checkout main && git pull
-git checkout -b hotfix/[short-description]
+BASE_BRANCH="$(git branch --show-current)" \
+    && git pull --ff-only \
+    && git checkout -b hotfix/[short-description] \
+    && git config branch."hotfix/[short-description]".trip-base "$BASE_BRANCH"
 ```
 
+If the active branch has no upstream, drop the `git pull --ff-only` step instead of ignoring its failure.
+
 ---
 
 ## Step 3: Minimal Investigation
@@ -106,12 +112,37 @@ git add -A && git commit -m "hotfix: [brief description]"
 
 ## Step 8: Merge & Tag
 
+Merge back into the base branch recorded in Step 2. Run as ONE command so the lookup and the merge share a shell:
+
+```bash
+BASE_BRANCH="$(git config branch.hotfix/[short-description].trip-base)" \
+    && git checkout "$BASE_BRANCH" \
+    && git merge --ff-only hotfix/[short-description] \
+    && git tag vx.y.z \
+    && git branch -d hotfix/[short-description]
+```
+
+If the `trip-base` lookup fails (branch created outside this skill), ask the user which branch to merge into. If `--ff-only` fails, the base branch moved during the fix — rebase the hotfix branch onto it, then retry. **Never create a merge commit.**
+
+## Step 8b: Push
+
+**If the user already specified which branch to push**, push that branch directly and skip the question.
+
+Otherwise, list the available branches:
+
+```bash
+git branch --format='%(refname:short)'
+```
+
+Then **use the `AskUserQuestion` tool** to ask:
+
+- **Question**: "Hotfix vx.y.z is merged and tagged. Which branch should I push?"
+- **Options**: One option per branch from the list above — the base branch first, marked "(recommended)" — plus "Not yet" (push manually later). If there are too many branches to list, include the most relevant ones; the user can name another via "Other".
+
+**If a branch is selected**:
+
 ```bash
-git checkout main
-git merge hotfix/[short-description]
-git tag vx.y.z
-git push && git push --tags
-git branch -d hotfix/[short-description]
+git push origin <selected-branch> && git push --tags
 ```
 
 ---
diff --git a/skills/TRIP-init/SKILL.md b/skills/TRIP-init/SKILL.md
index a8286aa..0bc9d57 100644
--- a/skills/TRIP-init/SKILL.md
+++ b/skills/TRIP-init/SKILL.md
@@ -645,7 +645,7 @@ Update Step 2 to reference the actual version file:
 
 **B. Week Anchor**
 
-The week Init is run becomes **Week 1** of the project. Capture the anchor date (Monday of the current week) and update the week formula in `TRIP-2-implement`.
+The week Init is run becomes **Week 1** of the project. Capture the anchor date (Monday of the current week) and update the week formula in `TRIP-3-release`.
 
 Run this to get the anchor date:
 
@@ -687,7 +687,7 @@ Then replace the `[WEEK_ANCHOR_DATE]` placeholder in `TRIP-3-release` Step 1 wit
 Then update the `[TUTORIAL_STEP]` block in `TRIP-3-release` with the user's context:
 
 ```markdown
-### Step 7: Tutorial
+### Step 8: Tutorial
 
 Create `docs/5-tuto/tuto_x.y.z.md` explaining the core principle.
 
@@ -705,6 +705,8 @@ Create `docs/5-tuto/tuto_x.y.z.md` explaining the core principle.
 - Step 8: README Update → **Step 9**: README Update
 - Step 9: Commit → **Step 10**: Commit
 - Step 10: Tag → **Step 11**: Tag
+- Step 11: Merge → **Step 12**: Merge
+- Step 12: Push → **Step 13**: Push
 
 **D. Codex Review Test Commands**
 
@@ -1012,10 +1014,9 @@ Update: Technology Stack, and any affected architectural sections
   - [ ] `TRIP-1-plan`: `[ADAPT_TO_PROJECT]` markers replaced with actual technical considerations
   - [ ] `TRIP-1-plan`: Guidance sections replaced with project-specific patterns
   - [ ] `TRIP-1-plan`: Custom plan sections added (if user requested)
-  - [ ] `TRIP-2-implement`: Testing gate commands (`[LINT_COMMAND]`, `[TYPECHECK_COMMAND]`, `[TEST_COMMAND]`) replaced with actual commands
+  - [ ] `TRIP-2-implement` and `TRIP-3-release`: Testing commands (`[LINT_COMMAND]`, `[TYPECHECK_COMMAND]`, `[TEST_COMMAND]`) replaced with actual commands
   - [ ] `TRIP-3-release`: `[VERSION_FILE]` placeholder replaced
   - [ ] `TRIP-3-release`: `[WEEK_ANCHOR_DATE]` placeholder replaced
-  - [ ] `TRIP-3-release`: Standalone-verification commands replaced with actual commands
   - [ ] `TRIP-3-release`: Tutorial preference configured (if enabled: 5-tuto/ folder created + user context; if disabled: `[TUTORIAL_STEP]` block removed)
   - [ ] `TRIP-review/checklist.md`: `[ADAPT_TO_PROJECT]` markers replaced with project-specific checklist sections
   - [ ] `TRIP-review/cr-template.md`: Checklist section names updated to match adapted `checklist.md`
diff --git a/skills/TRIP-research/SKILL.md b/skills/TRIP-research/SKILL.md
index 874fd83..50174f2 100644
--- a/skills/TRIP-research/SKILL.md
+++ b/skills/TRIP-research/SKILL.md
@@ -150,8 +150,7 @@ Conduct the research:
 For **decision-grade findings** — architecture recommendations, technology choices, anything the user will build on (typically compute level `think hard` and above) — red-team the draft conclusion with the `codex-ask` skill before presenting. Skip for quick lookups.
 
 ```bash
-export STATE_DIR=".claude/skills/codex-ask/state"
-bash .claude/skills/codex-plan-review/scripts/start.sh \
+bash .claude/skills/codex-ask/scripts/start.sh \
     --prompt-file .claude/skills/codex-ask/prompts/ask.tpl \
     <topic-label> "Here is my draft recommendation: <summary + key rationale>. Red-team it: what am I missing, what would you choose instead, and why?"
 ```
diff --git a/skills/TRIP-upgrade/SKILL.md b/skills/TRIP-upgrade/SKILL.md
index 80c2441..da1edfb 100644
--- a/skills/TRIP-upgrade/SKILL.md
+++ b/skills/TRIP-upgrade/SKILL.md
@@ -243,9 +243,8 @@ For each customized skill, take the **new template** from staging and inject the
 2. Replace `[PROJECT_NAME]` with extracted `PROJECT_NAME`
 3. Replace `[VERSION_FILE]` with extracted `VERSION_FILE`
 4. Replace `[WEEK_ANCHOR_DATE]` with extracted `WEEK_ANCHOR_DATE`
-5. Replace `[MAIN_BRANCH]` with the repo's default branch name
-6. Replace the standalone-verification commands with the same extracted lint/typecheck/test commands
-7. Handle tutorial config:
+5. Replace the standalone-verification commands with the same extracted lint/typecheck/test commands
+6. Handle tutorial config:
    - If tutorials were disabled: remove the `[TUTORIAL_STEP]` block
    - If tutorials were enabled: replace the `[TUTORIAL_STEP]` block with extracted `TUTORIAL_CONFIG` and renumber subsequent steps
 
@@ -281,7 +280,7 @@ After writing all files, run a validation pass.
 Scan all upgraded skill files for leftover placeholders:
 
 ```bash
-grep -rn '\[ADAPT_TO_PROJECT\|\[PROJECT_NAME\]\|\[VERSION_FILE\]\|\[WEEK_ANCHOR_DATE\]\|\[TEST_COMMAND\]\|\[LINT_COMMAND\]\|\[TYPECHECK_COMMAND\]\|\[TUTORIAL_STEP\]\|\[MAIN_BRANCH\]' .claude/skills/TRIP-*/
+grep -rn '\[ADAPT_TO_PROJECT\|\[PROJECT_NAME\]\|\[VERSION_FILE\]\|\[WEEK_ANCHOR_DATE\]\|\[TEST_COMMAND\]\|\[LINT_COMMAND\]\|\[TYPECHECK_COMMAND\]\|\[TUTORIAL_STEP\]\|\[MAIN_BRANCH\]' .claude/skills/
 ```
 
 If any are found, fill them from context or ask the user.
@@ -291,7 +290,7 @@ If any are found, fill them from context or ask the user.
 - `checklist.md` section names must match `cr-template.md` checklist section names
 - `codex-code-review/prompts/start.tpl` and `resume.tpl` reference `.claude/skills/TRIP-review/checklist.md` — confirm it exists, and that no template still points at the old `TRIP-3-review/` path
 - `codex-code-review/prompts/synthesize.tpl` and `codex-code-review/SKILL.md` reference `.claude/skills/TRIP-review/cr-template.md` — confirm it exists
-- `TRIP-1-plan` and `TRIP-2-implement` reference `codex-plan-review/scripts/start.sh` and `resume.sh`; `TRIP-2-implement` also references `codex-implement/scripts/start.sh` — confirm they exist
+- `TRIP-1-plan` references `codex-plan-review/scripts/`; `TRIP-2-implement` references the local wrappers under `codex-implement/scripts/` and `codex-code-review/scripts/` — confirm they exist
 
 ### 5.3 Present Summary
 
diff --git a/skills/codex-ask/SKILL.md b/skills/codex-ask/SKILL.md
index 221e52f..f59e5fb 100644
--- a/skills/codex-ask/SKILL.md
+++ b/skills/codex-ask/SKILL.md
@@ -10,11 +10,7 @@ Free-form second opinion from Codex CLI on **any matter** — architecture decis
 
 **Advisory, not authoritative.** Unlike `codex-plan-review` / `codex-code-review`, there are no verdict tags and nothing is gated on the answer: treat the response as one input to your judgment, exactly like a colleague's opinion. Agreement is weak evidence; *disagreement* is a strong signal that something deserves the user's attention.
 
-State persisted per topic label under `.claude/skills/codex-ask/state/` — follow-ups resume the same thread, enabling multi-round discussion. The shared scripts from `codex-plan-review` do all the work; always export first:
-
-```bash
-export STATE_DIR=".claude/skills/codex-ask/state"
-```
+State persists per topic label under `.claude/skills/codex-ask/state/`; follow-ups resume the same thread, enabling multi-round discussion. Local wrappers set the discussion state directory before delegating to the shared scripts.
 
 ## Arguments
 
@@ -26,19 +22,19 @@ export STATE_DIR=".claude/skills/codex-ask/state"
 
 1. **Start** a discussion:
    ```bash
-   bash .claude/skills/codex-plan-review/scripts/start.sh \
+   bash .claude/skills/codex-ask/scripts/start.sh \
        --prompt-file .claude/skills/codex-ask/prompts/ask.tpl \
        <topic-label> "<question — include your draft position and ask for disagreement>"
    ```
 
 2. **Follow up** in the same thread (counterpoints, new evidence):
    ```bash
-   bash .claude/skills/codex-plan-review/scripts/resume.sh \
+   bash .claude/skills/codex-ask/scripts/resume.sh \
        --prompt-file .claude/skills/codex-ask/prompts/followup.tpl \
        <topic-label> "<follow-up or counterpoint>"
    ```
 
-3. **Reset**: `bash .claude/skills/codex-plan-review/scripts/reset.sh <topic-label>` — **Show**: `show.sh <topic-label>`
+3. **Reset**: `bash .claude/skills/codex-ask/scripts/reset.sh <topic-label>` — **Show**: `bash .claude/skills/codex-ask/scripts/show.sh <topic-label>`
 
 ## When to use
 
diff --git a/skills/codex-ask/scripts/reset.sh b/skills/codex-ask/scripts/reset.sh
new file mode 100644
index 0000000..bdf0d0d
--- /dev/null
+++ b/skills/codex-ask/scripts/reset.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/reset.sh" "$@"
diff --git a/skills/codex-ask/scripts/resume.sh b/skills/codex-ask/scripts/resume.sh
new file mode 100644
index 0000000..67d3250
--- /dev/null
+++ b/skills/codex-ask/scripts/resume.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/resume.sh" "$@"
diff --git a/skills/codex-ask/scripts/show.sh b/skills/codex-ask/scripts/show.sh
new file mode 100644
index 0000000..a06e375
--- /dev/null
+++ b/skills/codex-ask/scripts/show.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/show.sh" "$@"
diff --git a/skills/codex-ask/scripts/start.sh b/skills/codex-ask/scripts/start.sh
new file mode 100644
index 0000000..25def94
--- /dev/null
+++ b/skills/codex-ask/scripts/start.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/start.sh" "$@"
diff --git a/skills/codex-ask/state/.gitignore b/skills/codex-ask/state/.gitignore
new file mode 100644
index 0000000..23e2664
--- /dev/null
+++ b/skills/codex-ask/state/.gitignore
@@ -0,0 +1,4 @@
+# Per-topic Codex discussion state (thread ids, answers, and event logs).
+# Local-only; these are session artifacts, not source files.
+*
+!.gitignore
diff --git a/skills/codex-code-review/SKILL.md b/skills/codex-code-review/SKILL.md
index 5dd6373..285132d 100644
--- a/skills/codex-code-review/SKILL.md
+++ b/skills/codex-code-review/SKILL.md
@@ -10,11 +10,7 @@ Iterative code review via Codex CLI on uncommitted changes. Codex reads the plan
 
 Review output stays in `state/<key>.review.txt` — not `docs/3-code-review/`. Promotion to `docs/3-code-review/CR_wa_vx.y.z.md` happens after convergence, not per-turn.
 
-State persisted under `.claude/skills/codex-code-review/state/<sanitized-target>.{thread,review.txt,events.ndjson}`. Shared scripts live under `.claude/skills/codex-plan-review/scripts/`; always export before invoking:
-
-```bash
-export STATE_DIR=".claude/skills/codex-code-review/state"
-```
+State persists under `.claude/skills/codex-code-review/state/<sanitized-target>.{thread,review.txt,events.ndjson}`. Local wrappers set the code-review state directory before delegating to the shared scripts.
 
 ## Arguments
 
@@ -27,12 +23,12 @@ export STATE_DIR=".claude/skills/codex-code-review/state"
 1. **Parse `$ARGUMENTS`**: extract action (`reset`/`show`/auto) and target.
 
 2. **Auto** — try `start.sh` first (exit code 2 = thread exists -> use `resume.sh`):
-   - **Start**: `bash .claude/skills/codex-plan-review/scripts/start.sh --prompt-file .claude/skills/codex-code-review/prompts/start.tpl <target> [extra]`
-   - **Resume**: `bash .claude/skills/codex-plan-review/scripts/resume.sh --prompt-file .claude/skills/codex-code-review/prompts/resume.tpl <target> [extra]`
+   - **Start**: `bash .claude/skills/codex-code-review/scripts/start.sh --prompt-file .claude/skills/codex-code-review/prompts/start.tpl <target> [extra]`
+   - **Resume**: `bash .claude/skills/codex-code-review/scripts/resume.sh --prompt-file .claude/skills/codex-code-review/prompts/resume.tpl <target> [extra]`
 
-3. **Reset**: `bash .claude/skills/codex-plan-review/scripts/reset.sh <target>`
+3. **Reset**: `bash .claude/skills/codex-code-review/scripts/reset.sh <target>`
 
-4. **Show**: `bash .claude/skills/codex-plan-review/scripts/show.sh <target>`
+4. **Show**: `bash .claude/skills/codex-code-review/scripts/show.sh <target>`
 
 5. **Parse trailing tag**:
    - `APPROVED` — propose post-convergence steps.
diff --git a/skills/codex-code-review/prompts/synthesize.tpl b/skills/codex-code-review/prompts/synthesize.tpl
index ef20e73..aedd805 100644
--- a/skills/codex-code-review/prompts/synthesize.tpl
+++ b/skills/codex-code-review/prompts/synthesize.tpl
@@ -9,7 +9,7 @@ Read `.claude/skills/TRIP-review/cr-template.md` and produce output conforming t
 Fill-in guide:
 - **Title**: feature/change name from `{{TARGET}}`
 - **Review Date**: today's date (YYYY-MM-DD)
-- **Version**: leave as `<x.y.z>` (requester fills from TRIP-2 Step 2)
+- **Version**: leave as `<x.y.z>` (requester fills from TRIP-3 Step 2)
 - **Files Reviewed**: from `git diff --name-only HEAD`
 - **Plan**: `{{TARGET}}` if path under `docs/1-plans/`, else "no plan — unplanned change"
 - **Findings**: every finding from all rounds with `file:line` and disposition
diff --git a/skills/codex-code-review/scripts/reset.sh b/skills/codex-code-review/scripts/reset.sh
new file mode 100644
index 0000000..bdf0d0d
--- /dev/null
+++ b/skills/codex-code-review/scripts/reset.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/reset.sh" "$@"
diff --git a/skills/codex-code-review/scripts/resume.sh b/skills/codex-code-review/scripts/resume.sh
new file mode 100644
index 0000000..67d3250
--- /dev/null
+++ b/skills/codex-code-review/scripts/resume.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/resume.sh" "$@"
diff --git a/skills/codex-code-review/scripts/show.sh b/skills/codex-code-review/scripts/show.sh
new file mode 100644
index 0000000..a06e375
--- /dev/null
+++ b/skills/codex-code-review/scripts/show.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/show.sh" "$@"
diff --git a/skills/codex-code-review/scripts/start.sh b/skills/codex-code-review/scripts/start.sh
new file mode 100644
index 0000000..25def94
--- /dev/null
+++ b/skills/codex-code-review/scripts/start.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/start.sh" "$@"
diff --git a/skills/codex-implement/SKILL.md b/skills/codex-implement/SKILL.md
index 2795161..7c1ab31 100644
--- a/skills/codex-implement/SKILL.md
+++ b/skills/codex-implement/SKILL.md
@@ -8,11 +8,7 @@ argument-hint: "<plan-path> [instructions] | reset <plan-path> | show <plan-path
 
 Non-interactive implementation via Codex CLI in a **workspace-write** sandbox: Codex reads the plan, edits the working tree directly, runs the project's lint/build on its own work, and reports back. One persistent thread per target, so multi-phase plans can be delegated phase by phase with full context retained.
 
-State persisted under `.claude/skills/codex-implement/state/<sanitized-target>.{thread,review.txt,events.ndjson}` (the `.review.txt` file holds Codex's implementation **report** — the naming comes from the shared helpers). `resume`/`reset`/`show` reuse the shared scripts from `codex-plan-review`; always export before invoking them:
-
-```bash
-export STATE_DIR=".claude/skills/codex-implement/state"
-```
+State persists under `.claude/skills/codex-implement/state/<sanitized-target>.{thread,review.txt,events.ndjson}` (the `.review.txt` file holds Codex's implementation **report**; the naming comes from the shared helpers). The local wrappers set the correct state directory before delegating to the shared scripts.
 
 ## Arguments
 
@@ -27,11 +23,11 @@ export STATE_DIR=".claude/skills/codex-implement/state"
 
 2. **Auto** — try `start.sh` first (exit code 2 = thread exists → use `resume.sh`):
    - **Start**: `bash .claude/skills/codex-implement/scripts/start.sh --prompt-file .claude/skills/codex-implement/prompts/implement.tpl <target> [instructions]`
-   - **Resume** (next phase / additional scope): `bash .claude/skills/codex-plan-review/scripts/resume.sh --prompt-file .claude/skills/codex-implement/prompts/continue.tpl <target> [instructions]`
+   - **Resume** (next phase / additional scope): `bash .claude/skills/codex-implement/scripts/resume.sh --prompt-file .claude/skills/codex-implement/prompts/continue.tpl <target> [instructions]`
 
-3. **Reset**: `bash .claude/skills/codex-plan-review/scripts/reset.sh <target>`
+3. **Reset**: `bash .claude/skills/codex-implement/scripts/reset.sh <target>`
 
-4. **Show**: `bash .claude/skills/codex-plan-review/scripts/show.sh <target>`
+4. **Show**: `bash .claude/skills/codex-implement/scripts/show.sh <target>`
 
 5. **Parse trailing tag** of the report:
    - `IMPLEMENTATION_COMPLETE` — hand control back to the requester's self-review (TRIP-2).
diff --git a/skills/codex-implement/scripts/reset.sh b/skills/codex-implement/scripts/reset.sh
new file mode 100644
index 0000000..bdf0d0d
--- /dev/null
+++ b/skills/codex-implement/scripts/reset.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/reset.sh" "$@"
diff --git a/skills/codex-implement/scripts/resume.sh b/skills/codex-implement/scripts/resume.sh
new file mode 100644
index 0000000..67d3250
--- /dev/null
+++ b/skills/codex-implement/scripts/resume.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/resume.sh" "$@"
diff --git a/skills/codex-implement/scripts/show.sh b/skills/codex-implement/scripts/show.sh
new file mode 100644
index 0000000..a06e375
--- /dev/null
+++ b/skills/codex-implement/scripts/show.sh
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+export STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
+exec bash "$SCRIPT_DIR/../../codex-plan-review/scripts/show.sh" "$@"
diff --git a/skills/codex-implement/scripts/start.sh b/skills/codex-implement/scripts/start.sh
old mode 100755
new mode 100644
index 4e6f1af..202b4dc
--- a/skills/codex-implement/scripts/start.sh
+++ b/skills/codex-implement/scripts/start.sh
@@ -14,9 +14,10 @@
 
 set -euo pipefail
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-# Default state to THIS skill's directory (shared _common.sh would
-# otherwise default to codex-plan-review's state).
-STATE_DIR="${STATE_DIR:-$(cd "$SCRIPT_DIR/.." && pwd)/state}"
+# Pin state to THIS skill's directory, like the wrapper scripts do —
+# unconditionally, so a stale STATE_DIR in the environment can't
+# redirect implementation state (or flip the model selection).
+STATE_DIR="$(cd "$SCRIPT_DIR/.." && pwd)/state"
 export STATE_DIR
 # shellcheck source=../../codex-plan-review/scripts/_common.sh
 source "$SCRIPT_DIR/../../codex-plan-review/scripts/_common.sh"
diff --git a/skills/codex-implement/state/.gitignore b/skills/codex-implement/state/.gitignore
new file mode 100644
index 0000000..8865ae4
--- /dev/null
+++ b/skills/codex-implement/state/.gitignore
@@ -0,0 +1,4 @@
+# Per-target Codex implementation state (thread ids, reports, and event logs).
+# Local-only; these are session artifacts, not source files.
+*
+!.gitignore
diff --git a/skills/codex-plan-review/scripts/_common.sh b/skills/codex-plan-review/scripts/_common.sh
old mode 100755
new mode 100644
index d59807e..998ded5
--- a/skills/codex-plan-review/scripts/_common.sh
+++ b/skills/codex-plan-review/scripts/_common.sh
@@ -5,8 +5,7 @@
 set -euo pipefail
 
 SKILL_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-# STATE_DIR can be overridden by the caller (e.g., codex-code-review
-# exports its own state path before invoking the shared scripts).
+# STATE_DIR can be overridden by a consumer wrapper (e.g., codex-code-review).
 # Default falls back to the script's own skill directory.
 : "${STATE_DIR:=$SKILL_DIR/state}"
 export STATE_DIR
@@ -68,12 +67,20 @@ load_prompt() {
         echo "error: prompt template not found: $tpl" >&2
         return 1
     fi
-    awk -v target="${TARGET-}" -v extra="${EXTRA_PROMPT-}" -v notes="${IMPLEMENTER_NOTES-}" '
+    awk '
+        function replace_literal(text, needle, replacement,    out, position) {
+            out = ""
+            while ((position = index(text, needle)) != 0) {
+                out = out substr(text, 1, position - 1) replacement
+                text = substr(text, position + length(needle))
+            }
+            return out text
+        }
         {
-            gsub(/\{\{TARGET\}\}/, target)
-            gsub(/\{\{EXTRA_PROMPT\}\}/, extra)
-            gsub(/\{\{IMPLEMENTER_NOTES\}\}/, notes)
-            print
+            line = replace_literal($0, "{{TARGET}}", ENVIRON["TARGET"])
+            line = replace_literal(line, "{{EXTRA_PROMPT}}", ENVIRON["EXTRA_PROMPT"])
+            line = replace_literal(line, "{{IMPLEMENTER_NOTES}}", ENVIRON["IMPLEMENTER_NOTES"])
+            print line
         }
     ' "$tpl"
 }
diff --git a/skills/codex-plan-review/scripts/key.sh b/skills/codex-plan-review/scripts/key.sh
new file mode 100644
index 0000000..b5bd977
--- /dev/null
+++ b/skills/codex-plan-review/scripts/key.sh
@@ -0,0 +1,13 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=_common.sh
+source "$SCRIPT_DIR/_common.sh"
+
+if [ $# -ne 1 ]; then
+    echo "usage: key.sh <target>" >&2
+    exit 64
+fi
+
+target_key "$1"
diff --git a/skills/codex-plan-review/scripts/reset.sh b/skills/codex-plan-review/scripts/reset.sh
old mode 100755
new mode 100644
diff --git a/skills/codex-plan-review/scripts/resume.sh b/skills/codex-plan-review/scripts/resume.sh
old mode 100755
new mode 100644
diff --git a/skills/codex-plan-review/scripts/show.sh b/skills/codex-plan-review/scripts/show.sh
old mode 100755
new mode 100644
diff --git a/skills/codex-plan-review/scripts/start.sh b/skills/codex-plan-review/scripts/start.sh
old mode 100755
new mode 100644
diff --git a/tests/workflow-regression.sh b/tests/workflow-regression.sh
new file mode 100644
index 0000000..b761c3d
--- /dev/null
+++ b/tests/workflow-regression.sh
@@ -0,0 +1,104 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+TMP_DIR="$(mktemp -d)"
+trap 'rm -rf "$TMP_DIR"' EXIT
+
+fail() {
+    echo "FAIL: $*" >&2
+    exit 1
+}
+
+assert_eq() {
+    local expected="$1"
+    local actual="$2"
+    local label="$3"
+    if [ "$actual" != "$expected" ]; then
+        printf 'FAIL: %s\nexpected: %s\nactual:   %s\n' "$label" "$expected" "$actual" >&2
+        exit 1
+    fi
+}
+
+test_prompt_substitution_preserves_literal_text() {
+    local template="$TMP_DIR/prompt.tpl"
+    printf 'Target={{TARGET}}\nExtra={{EXTRA_PROMPT}}\nNotes={{IMPLEMENTER_NOTES}}\n' > "$template"
+
+    export STATE_DIR="$TMP_DIR/state"
+    export TARGET='docs/C:\work\plan & notes.md'
+    export EXTRA_PROMPT='Keep A & B; preserve C:\temp\file.'
+    export IMPLEMENTER_NOTES='Fixed X & Y; left D:\repo\case unchanged.'
+
+    # shellcheck source=../skills/codex-plan-review/scripts/_common.sh
+    source "$ROOT/skills/codex-plan-review/scripts/_common.sh"
+    local actual
+    actual="$(load_prompt "$template")"
+    local expected
+    expected=$'Target=docs/C:\\work\\plan & notes.md\nExtra=Keep A & B; preserve C:\\temp\\file.\nNotes=Fixed X & Y; left D:\\repo\\case unchanged.'
+    assert_eq "$expected" "$actual" "prompt substitution must preserve ampersands and backslashes"
+}
+
+test_prompt_substitution_terminates_when_value_quotes_placeholder() {
+    local template="$TMP_DIR/self-ref.tpl"
+    local runner="$TMP_DIR/self-ref-runner.sh"
+    printf 'Notes={{IMPLEMENTER_NOTES}}\n' > "$template"
+
+    # Run in a subprocess under timeout: a regression here is an infinite
+    # loop, which must fail the test rather than hang the suite.
+    cat > "$runner" <<EOF
+export STATE_DIR="$TMP_DIR/state"
+export TARGET='x' EXTRA_PROMPT=''
+export IMPLEMENTER_NOTES='fixed the {{IMPLEMENTER_NOTES}} handling'
+source "$ROOT/skills/codex-plan-review/scripts/_common.sh"
+load_prompt "$template"
+EOF
+
+    local actual
+    actual="$(timeout 10 bash "$runner")" \
+        || fail "substitution hung or errored when the value quotes its own placeholder"
+    assert_eq 'Notes=fixed the {{IMPLEMENTER_NOTES}} handling' "$actual" \
+        "quoted placeholder in a value must pass through literally, not recurse"
+}
+
+test_key_helper_matches_shared_target_key() {
+    local actual
+    actual="$(bash "$ROOT/skills/codex-plan-review/scripts/key.sh" 'unplanned review & fix')"
+    assert_eq 'unplanned_review___fix' "$actual" "free-form labels must use the shared key derivation"
+}
+
+test_consumer_show_wrappers_use_their_own_state() {
+    # Run against a copy of the skills tree so fixtures never land in the
+    # real state directories, even when an assertion fails mid-loop.
+    local sandbox="$TMP_DIR/wrapper-sandbox"
+    mkdir -p "$sandbox"
+    cp -r "$ROOT/skills" "$sandbox/"
+
+    local skill label state output
+    for skill in codex-implement codex-code-review codex-ask; do
+        label="wrapper-test-$skill"
+        state="$sandbox/skills/$skill/state"
+        mkdir -p "$state"
+        printf 'thread-%s\n' "$skill" > "$state/$label.thread"
+        printf 'report-%s\n' "$skill" > "$state/$label.review.txt"
+        output="$(bash "$sandbox/skills/$skill/scripts/show.sh" "$label")"
+        grep -q "report-$skill" <<< "$output" || fail "$skill show wrapper read the wrong state directory"
+    done
+}
+
+test_runtime_state_is_ignored() {
+    local skill
+    for skill in codex-implement codex-ask; do
+        [ -f "$ROOT/skills/$skill/state/.gitignore" ] || fail "$skill state directory needs a .gitignore"
+        grep -qx '\*' "$ROOT/skills/$skill/state/.gitignore" || fail "$skill state .gitignore must ignore runtime files"
+        grep -qx '!\.gitignore' "$ROOT/skills/$skill/state/.gitignore" || fail "$skill state .gitignore must keep itself"
+    done
+}
+
+test_prompt_substitution_preserves_literal_text
+test_prompt_substitution_terminates_when_value_quotes_placeholder
+test_key_helper_matches_shared_target_key
+test_consumer_show_wrappers_use_their_own_state
+test_runtime_state_is_ignored
+
+echo "workflow regression tests passed"