claude-toolkit 0.10.0 → 0.15.0

package/CHANGELOG.md

@@ -1,5 +1,26 @@
 # Changelog
 
+## 0.15.0 (2026-07-04)
+
+Two new authoring-convention assets — an opt-in `esnext` stack and an always-on `ct-code-style` core skill — plus Biome enforcement of the mechanically-checkable style rules.
+
+- feat: add the `esnext` stack (skill `ct-esnext-idioms`) — availability-aware guidance on modern ECMAScript runtime features: `Temporal` over `Date`, `using`/`await using` for deterministic cleanup, lazy iterator helpers, `structuredClone`, `Object.groupBy`/`Map.groupBy`, `Promise.withResolvers`, immutable array ops, and ESM hygiene. Detected on tsconfig `target`/`module`/`lib` = ESNext/ES2022+, package.json `"type":"module"`, `engines.node >= 20`, or a `.mjs` file.
+- feat: add the `ct-code-style` core skill (always included) — universal code-structure conventions: guard clauses over nested `else`, object lookup tables over `if/else if` chains, single-level ternaries, full ES6 destructuring, and arrow-only functions with no `class`/`this`. Carve-outs defer to stack skills: never destructure reactive SolidJS `props`, and use `class`/`this` only where a platform mandates it (e.g. Cloudflare Durable Objects).
+- feat: enforce the mechanically-checkable style rules in the scaffolded Biome config — `noUselessElse`, `noNestedTernary`, and `useArrowFunction`. `noMagicNumbers` is kept as skill guidance only (too noisy to enforce).
+- docs: document both assets in the README, `docs/stacks/esnext-idioms.md`, `docs/skills/code-style.md`, and the docs index; cross-link the SolidJS props rule to `ct-code-style`.
+
+## 0.11.0 (2026-07-04)
+
+Remediation of the 2026-07-04 core audit (46 findings): the skill-eval hook is hardened, stack-owned commands are gated behind their stack, and the `ct-` agents, commands, skills, and reference docs are corrected and aligned.
+
+- feat: gate stack commands behind their owning stack — `proto-check` moved to the protobuf stack and is only installed when protobuf is detected. The skill hook is now registered as a single command string rather than exec-form args, so it runs correctly on every prompt.
+- fix(hooks): harden the skill-eval hook — correct glob-to-regex conversion, robust file-path extraction with an input cap, defensive config coercion with defaults and clean exit, output sanitization, and verification that referenced skills actually ship. Dropped ghost rules and locked the skill-rules schema with `additionalProperties: false`. Renamed to `.cjs` and exposed a testable module API.
+- fix(agents): register `ct-code-reviewer` and `ct-github-workflow` under their `ct-` names with scoped tool grants; give `ct-github-workflow` a real merge-conflict probe and default-branch resolution.
+- fix(commands): scope `Bash` grants and add `argument-hint` across `code-quality`, `onboard`, `ticket`, and `pr-summary`; `pr-review` now fetches the PR diff and delegates to `ct-code-reviewer`.
+- fix(skills): correct the `satisfies` example, align type-vs-interface guidance with best-practices, narrow E2E scope, and add edge cases plus "Use-when" triggers to the debugging and verification skills.
+- docs: add `CLAUDE.md` architecture guide, regenerate testing-patterns from the current skill, correct command invocation names to `/ct:`, and mirror the doc references to the moved/renamed assets.
+- chore: add `.gitattributes` to normalize line endings to LF across platforms.
+
 ## 0.10.0 (2026-06-09)
 
 One idempotent CLI command replaces `init`/`update`/`sync`, and `.claude/` now regenerates itself on toolkit upgrade.

package/README.md

@@ -107,6 +107,7 @@ Stacks in your config that are no longer detected are reported but left unchange
 | `playwright`      | Playwright E2E testing, Page Objects, fixtures, CI/CD              |
 | `storybook`       | Storybook interaction testing, CSF 3, visual regression            |
 | `capacitor`       | Capacitor 8 runtime, Capgo OTA, channels; webview UI & native feel |
+| `esnext`          | Modern ECMAScript idioms — Temporal, iterator helpers, `using`, structuredClone |
 
 ## Core Features (always included)
 
@@ -122,6 +123,7 @@ Stacks in your config that are no longer detected are reported but left unchange
 - `ct-testing-patterns` — TDD workflow and patterns
 - `ct-typescript-conventions` — TypeScript strict mode best practices
 - `ct-verification-before-completion` — Evidence-based completion claims
+- `ct-code-style` — Code structure & style: guard clauses, lookup tables, no magic values
 
 ### Core Commands
 

package/core/agents/ct-code-reviewer.md

@@ -1,7 +1,8 @@
 ---
-name: "Code Reviewer"
-description: "Senior code reviewer with stack-aware analysis"
+name: ct-code-reviewer
+description: Senior code reviewer with stack-aware analysis
 model: opus
+tools: Read, Grep, Glob, Bash
 ---
 
 # Code Reviewer Agent
@@ -75,7 +76,7 @@ You are a senior code reviewer. Your job is to review code changes thoroughly, p
 
 ## Output Format
 
-```markdown
+````markdown
 ## Code Review
 
 **Reviewed:** {files or PR description}
@@ -112,7 +113,7 @@ You are a senior code reviewer. Your job is to review code changes thoroughly, p
 
 ### Summary
 {1-3 sentence overall assessment. Be direct.}
-```
+````
 
 ## Principles
 

package/core/agents/ct-github-workflow.md

@@ -1,6 +1,6 @@
 ---
-name: "GitHub Workflow"
-description: "Git workflow assistant for branching, commits, and PRs"
+name: ct-github-workflow
+description: Git workflow assistant for branching, commits, and PRs
 model: sonnet
 ---
 
@@ -64,18 +64,27 @@ perf(search): add database index for full-text queries
 ## PR Creation Workflow
 
 ### 1. Pre-flight Checks
-```bash
-# Verify tests pass
-{project test command}
 
-# Verify types check
-{project typecheck command}
+First run the project's own verification commands (substitute this project's actual scripts — these are placeholders, not literal commands):
+- **Tests** — run the project's test command
+- **Types** — run the project's typecheck command
+- **Lint** — run the project's lint command
 
-# Verify no lint errors
-{project lint command}
+Then confirm the branch is current with, and conflict-free against, the default branch:
 
-# Check for merge conflicts
-git fetch origin main && git merge-base --is-ancestor origin/main HEAD
+```bash
+# Resolve the default branch (main, master, ...) instead of assuming main
+BASE=$(git symbolic-ref --short refs/remotes/origin/HEAD 2>/dev/null | sed 's@^origin/@@')
+BASE=${BASE:-main}
+git fetch origin "$BASE"
+
+# Up-to-date check: a non-zero exit means the base has commits not yet on this
+# branch (i.e. the branch is behind) — this is NOT a conflict signal.
+git merge-base --is-ancestor "origin/$BASE" HEAD
+
+# Real merge-conflict probe (does not touch the working tree): a non-zero exit
+# means merging the base into this branch would produce conflicts.
+git merge-tree --write-tree "origin/$BASE" HEAD >/dev/null
 ```
 
 ### 2. PR Title
@@ -114,8 +123,12 @@ Keep under 70 characters. The title should tell a reviewer what this PR does wit
 ## Workflow Commands
 
 ```bash
-# Create feature branch from latest main
-git checkout main && git pull && git checkout -b feat/description
+# Resolve the default branch once (main, master, ...)
+BASE=$(git symbolic-ref --short refs/remotes/origin/HEAD 2>/dev/null | sed 's@^origin/@@')
+BASE=${BASE:-main}
+
+# Create feature branch from the latest default branch
+git checkout "$BASE" && git pull && git checkout -b feat/description
 
 # Stage and commit with conventional message
 git add {files}
@@ -125,8 +138,8 @@ git commit -m "feat(scope): description"
 git push -u origin HEAD
 gh pr create --title "feat(scope): description" --body "..."
 
-# Update branch with latest main
-git fetch origin main && git rebase origin/main
+# Update branch with the latest default branch
+git fetch origin "$BASE" && git rebase "origin/$BASE"
 ```
 
 ## Principles

package/core/commands/ct/code-quality.md

@@ -1,7 +1,25 @@
 ---
-description: "Run code quality checks on a directory or file"
+description: Run code quality checks on a directory or file
+argument-hint: "[path]"
 allowed-tools:
-  - Bash
+  - Bash(eslint:*)
+  - Bash(tsc:*)
+  - Bash(prettier:*)
+  - Bash(biome:*)
+  - Bash(ruff:*)
+  - Bash(mypy:*)
+  - Bash(pyright:*)
+  - Bash(black:*)
+  - Bash(cargo clippy:*)
+  - Bash(cargo check:*)
+  - Bash(rustfmt:*)
+  - Bash(gofmt:*)
+  - Bash(go vet:*)
+  - Bash(golangci-lint:*)
+  - Bash(npm run:*)
+  - Bash(pnpm:*)
+  - Bash(bun run:*)
+  - Bash(yarn:*)
   - Read
   - Glob
   - Grep
@@ -11,6 +29,8 @@ allowed-tools:
 
 Run lint, typecheck, and format checks against the target directory or file. Report all issues found and suggest fixes.
 
+**Target:** `$ARGUMENTS` (defaults to the current directory if empty).
+
 ## Workflow
 
 1. **Detect project tooling** by reading `package.json`, `pyproject.toml`, `Makefile`, `Cargo.toml`, or equivalent config files in the project root. Identify which linter, typechecker, and formatter are configured.

package/core/commands/ct/onboard.md

@@ -1,7 +1,12 @@
 ---
-description: "Onboard yourself to a codebase area or task"
+description: Onboard yourself to a codebase area or task
+argument-hint: "[area]"
 allowed-tools:
-  - Bash
+  - Bash(git log:*)
+  - Bash(git ls-files:*)
+  - Bash(git remote:*)
+  - Bash(ls:*)
+  - Bash(find:*)
   - Read
   - Glob
   - Grep
@@ -13,7 +18,7 @@ Systematically explore and understand a codebase area, module, or task context.
 
 ## Workflow
 
-1. **Understand the scope.** The user will specify an area (e.g., "the auth module", "the payments flow", "this repo"). If no scope is given, start from the project root.
+1. **Understand the scope.** The scope is given by `$ARGUMENTS` (e.g., "the auth module", "the payments flow", "this repo"). If `$ARGUMENTS` is empty, start from the project root.
 
 2. **Read foundational files first:**
    - `README.md`, `CLAUDE.md`, `CONTRIBUTING.md` (project conventions)

package/core/commands/ct/pr-review.md

@@ -1,104 +1,38 @@
 ---
-description: "Review a pull request using project standards"
+description: Review a pull request using project standards
+argument-hint: "[pr-number]"
 allowed-tools:
-  - Bash
+  - Bash(gh pr view:*)
+  - Bash(gh pr diff:*)
+  - Bash(gh pr checkout:*)
+  - Bash(git diff:*)
+  - Bash(git log:*)
+  - Bash(git merge-base:*)
+  - Bash(git branch:*)
   - Read
-  - Glob
-  - Grep
+  - Task
 ---
 
 # Pull Request Review
 
-Perform a checklist-based code review on the current PR or a specified PR number. Evaluate against project standards and common quality criteria.
+Review the current branch's PR, or a specific PR number passed as `$ARGUMENTS`, by delegating the analysis to the `ct-code-reviewer` agent — the single source of truth for the review checklist, severity model, and output format.
 
 ## Workflow
 
-1. **Get PR context.** If a PR number is provided, fetch it with `gh pr view {number} --json title,body,files,commits`. Otherwise, diff the current branch against the base branch (`main` or `master`).
+1. **Get PR context.**
+   - If `$ARGUMENTS` contains a PR number, fetch the PR's **actual** changes (metadata alone has no patch):
+     - `gh pr view $ARGUMENTS --json title,body,files,commits` for the title, description, and file list.
+     - `gh pr diff $ARGUMENTS` to read the actual patch.
+     - If you need to read changed files at the PR's revision rather than the local checkout, run `gh pr checkout $ARGUMENTS` first so the working tree matches the PR branch.
+   - Otherwise (no PR number), diff the current branch against its base branch (`main` or `master`): resolve the base with `git merge-base`, then use `git diff {base}..HEAD` and `git log {base}..HEAD --oneline --no-merges`.
 
-2. **Read all changed files** in full to understand the complete context of the change. Do not review diffs in isolation — understand what the surrounding code does.
+2. **Delegate the review to the `ct-code-reviewer` agent.** Spawn the `ct-code-reviewer` agent with the Task tool, handing it the PR title/description and the diff (and, when you ran `gh pr checkout`, permission to read the changed files in full). The agent applies its own review checklist, severity model, and output format — do not restate them here.
 
-3. **Run the review checklist** against every changed file.
-
-4. **Compile findings** into a structured review.
-
-## Review Checklist
-
-### Logic Correctness
-- Are conditionals and branches correct? Any off-by-one errors?
-- Are edge cases handled (empty inputs, null values, boundary conditions)?
-- Does the control flow match the intended behavior?
-- Are there any unreachable code paths?
-
-### Type Safety
-- Are types precise (no `any`, no unsafe casts, no overly broad unions)?
-- Is type narrowing used correctly before accessing properties?
-- Are generic constraints appropriate?
-- Are return types explicit where they should be?
-
-### Error Handling
-- Are errors caught and handled meaningfully (not swallowed)?
-- Do error messages provide enough context for debugging?
-- Are error boundaries or fallbacks in place where needed?
-- Is cleanup logic (finally blocks, defer, etc.) correct?
-
-### Test Coverage
-- Are new behaviors covered by tests?
-- Do tests focus on behavior, not implementation details?
-- Are edge cases and error paths tested?
-- Are test descriptions clear and descriptive?
-
-### Performance
-- Any unnecessary re-renders, recomputations, or allocations?
-- Any N+1 query patterns or missing pagination?
-- Any missing memoization for expensive operations?
-- Any potential memory leaks (unclosed resources, dangling listeners)?
-
-### Security
-- Any injection vulnerabilities (SQL, XSS, command injection)?
-- Are user inputs validated and sanitized?
-- Are auth checks present on protected operations?
-- Are secrets or sensitive data properly handled (not logged, not exposed)?
-
-### Code Style
-- Does naming follow project conventions?
-- Is the code readable without excessive comments?
-- Are abstractions at the right level (not over- or under-engineered)?
-- Is there unnecessary duplication?
-
-## Output Format
-
-```markdown
-## PR Review: {title}
-
-**Branch:** {branch} -> {base}
-**Files changed:** {count}
-**Verdict:** {APPROVE | REQUEST_CHANGES | COMMENT}
-
-### Critical Issues
-> Must be fixed before merge.
-
-- **[CRITICAL]** {file}:{line} - {description}
-
-### Warnings
-> Should be addressed, but not blocking.
-
-- **[WARNING]** {file}:{line} - {description}
-
-### Suggestions
-> Nice-to-have improvements.
-
-- **[SUGGESTION]** {file}:{line} - {description}
-
-### Positive Notes
-- {things done well}
-
-### Summary
-{1-3 sentence overall assessment}
-```
+3. **Return the agent's review to the user verbatim.** Add a one-line note only if you had to skip something (e.g. the PR was too large to check out, or `gh` was unavailable).
 
 ## Notes
 
-- Be specific. Reference file names and line numbers.
-- Distinguish between opinion and objective issues.
-- Acknowledge good patterns, not just problems.
+- The `ct-code-reviewer` agent is the canonical review checklist and output format. This command only gathers context and delegates, so the checklist and severity model never drift between the two.
+- Prefer `gh pr diff` / `gh pr checkout` over reading local files when a PR number is given — the local branch is usually not the PR branch.
 - If the PR is too large to review effectively, say so and suggest splitting it.
+- If `gh` is not installed or the PR number is invalid, fall back to reviewing the current branch's diff against its base.

package/core/commands/ct/pr-summary.md

@@ -1,7 +1,10 @@
 ---
-description: "Generate a PR summary from the current branch"
+description: Generate a PR summary from the current branch
 allowed-tools:
-  - Bash
+  - Bash(git diff:*)
+  - Bash(git log:*)
+  - Bash(git merge-base:*)
+  - Bash(git branch:*)
   - Read
 ---
 
@@ -15,7 +18,7 @@ Generate a structured pull request summary by analyzing the diff and commit hist
 
 2. **Gather the diff.** Run `git diff {base}...HEAD --stat` for an overview and `git diff {base}...HEAD` for full changes.
 
-3. **Gather commit history.** Run `git log {base}...HEAD --oneline --no-merges` to understand the progression of changes.
+3. **Gather commit history.** Run `git log {base}..HEAD --oneline --no-merges` (two-dot: commits on HEAD but not on base) to understand the progression of changes.
 
 4. **Analyze changes** by reading modified files to understand the "why" behind each change. Group related changes together.
 

package/core/commands/ct/ticket.md

@@ -1,7 +1,24 @@
 ---
-description: "Work end-to-end on a ticket or issue"
+description: Work end-to-end on a ticket or issue
+argument-hint: "[issue-number | description]"
 allowed-tools:
-  - Bash
+  - Bash(git status:*)
+  - Bash(git checkout:*)
+  - Bash(git switch:*)
+  - Bash(git branch:*)
+  - Bash(git add:*)
+  - Bash(git commit:*)
+  - Bash(git push:*)
+  - Bash(git diff:*)
+  - Bash(git log:*)
+  - Bash(gh issue view:*)
+  - Bash(gh pr create:*)
+  - Bash(gh pr view:*)
+  - Bash(npm run:*)
+  - Bash(pnpm:*)
+  - Bash(bun run:*)
+  - Bash(bun test:*)
+  - Bash(yarn:*)
   - Read
   - Write
   - Edit
@@ -17,7 +34,7 @@ Work on a ticket or issue end-to-end: understand it, explore the codebase, plan
 
 ### Phase 1: Understand
 
-1. **Read the ticket.** If a GitHub issue number is provided, fetch it with `gh issue view {number}`. Otherwise, use the description provided by the user.
+1. **Read the ticket.** If `$ARGUMENTS` is a GitHub issue number, fetch it with `gh issue view $ARGUMENTS`. Otherwise, treat `$ARGUMENTS` as the ticket description.
 
 2. **Clarify requirements.** Identify acceptance criteria, edge cases, and constraints. If anything is ambiguous, ask before proceeding.
 

package/core/hooks/{skill-eval.js → skill-eval.cjs}

@@ -1,4 +1,3 @@
-#!/usr/bin/env node
 /**
  * Skill Evaluation Engine v2.0
  *
@@ -18,6 +17,10 @@ const path = require("node:path");
 // Configuration
 const RULES_PATH = path.join(__dirname, "skill-rules.json");
 
+// Cap the text handed to the (potentially quadratic) regex scans. Long single-line
+// pastes — minified bundles, one-line JSON — can otherwise blow past the 5s hook timeout.
+const MAX_PROMPT_LENGTH = 50_000;
+
 /**
  * @typedef {Object} SkillMatch
  * @property {string} name - Skill name
@@ -36,7 +39,7 @@ function loadRules() {
 		return JSON.parse(content);
 	} catch (error) {
 		console.error(`Failed to load skill rules: ${error.message}`);
-		process.exit(0);
+		return null;
 	}
 }
 
@@ -46,29 +49,33 @@ function loadRules() {
  * @returns {string[]} Array of detected file paths
  */
 function extractFilePaths(prompt) {
+	// Normalize backslashes so Windows paths match, and cap length for the regex scans.
+	const text = String(prompt).slice(0, MAX_PROMPT_LENGTH).replace(/\\/g, "/");
 	const paths = new Set();
 
-	// Match explicit paths with extensions
+	// Match explicit paths with extensions. Negative lookbehind lets a candidate start
+	// after any non-path delimiter — "(", ",", ":", "=", whitespace, quotes, or BOL.
 	const extensionPattern =
-		/(?:^|\s|["'`])([\w\-./]+\.(?:[tj]sx?|json|gql|ya?ml|md|sh|rs|proto|toml|css\.ts|wasm))\b/gi;
-	for (const match of prompt.matchAll(extensionPattern)) {
+		/(?<![\w\-./])([\w\-./]+\.(?:[tj]sx?|json|gql|ya?ml|md|sh|rs|proto|toml|wasm))\b/gi;
+	for (const match of text.matchAll(extensionPattern)) {
 		paths.add(match[1]);
 	}
 
-	// Match paths starting with common directories
+	// Match paths starting with common directories.
 	const dirPattern =
-		/(?:^|\s|["'`])((?:src|app|components|worker|workers|proto|hooks|utils|services|styles|locales|tests|\.claude|\.github)\/[\w\-./]+)/gi;
-	for (const match of prompt.matchAll(dirPattern)) {
+		/(?<![\w\-./])((?:src|app|components|worker|workers|proto|hooks|utils|services|styles|locales|tests|\.claude|\.github)\/[\w\-./]+)/gi;
+	for (const match of text.matchAll(dirPattern)) {
 		paths.add(match[1]);
 	}
 
-	// Match quoted paths
+	// Match quoted paths.
 	const quotedPattern = /["'`]([\w\-./]+\/[\w\-./]+)["'`]/g;
-	for (const match of prompt.matchAll(quotedPattern)) {
+	for (const match of text.matchAll(quotedPattern)) {
 		paths.add(match[1]);
 	}
 
-	return Array.from(paths);
+	// Require at least one letter so ratios/fractions like "3/4" are not treated as paths.
+	return Array.from(paths).filter((p) => /[A-Za-z]/.test(p));
 }
 
 /**
@@ -94,18 +101,19 @@ function matchesPattern(text, pattern, flags = "i") {
  * @returns {boolean}
  */
 function matchesGlob(filePath, globPattern) {
+	const normalized = String(filePath).replace(/\\/g, "/");
 	const regexPattern = globPattern
-		.replace(/\./g, "\\.")
-		.replace(/\*\*\//g, "<<<DOUBLESTARSLASH>>>")
-		.replace(/\*\*/g, "<<<DOUBLESTAR>>>")
+		.replace(/[.+^${}()|[\]]/g, "\\$&")
+		.replace(/\*\*\//g, "<<<GLOBSTARSLASH>>>")
+		.replace(/\*\*/g, "<<<GLOBSTAR>>>")
 		.replace(/\*/g, "[^/]*")
-		.replace(/<<<DOUBLESTARSLASH>>>/g, "(.*\\/)?")
-		.replace(/<<<DOUBLESTAR>>>/g, ".*")
-		.replace(/\?/g, ".");
+		.replace(/\?/g, ".")
+		.replace(/<<<GLOBSTARSLASH>>>/g, "(?:.*/)?")
+		.replace(/<<<GLOBSTAR>>>/g, ".*");
 
 	try {
 		const regex = new RegExp(`^${regexPattern}$`, "i");
-		return regex.test(filePath);
+		return regex.test(normalized);
 	} catch {
 		return false;
 	}
@@ -257,20 +265,48 @@ function formatConfidence(score, minScore) {
 	return "LOW";
 }
 
+// Strip characters that could break out of the <user-prompt-submit-hook> block or
+// forge system-reminder content when a rule-derived string is echoed into output (F04).
+function sanitize(value) {
+	return String(value)
+		.replace(/[<>\r\n]/g, " ")
+		.trim();
+}
+
+// Confirm the skill actually ships before ordering Claude to invoke it (F03).
+// baseDir is .claude/skills at runtime (path.join(__dirname, "..", "skills")).
+function skillExists(name, skillsDir) {
+	try {
+		return fs.existsSync(path.join(skillsDir, name, "SKILL.md"));
+	} catch {
+		return false;
+	}
+}
+
 /**
  * Main evaluation function
  */
-function evaluate(prompt) {
-	const rules = loadRules();
-	const { config, skills } = rules;
-
-	const promptLower = prompt.toLowerCase();
-	const filePaths = extractFilePaths(prompt);
+function evaluate(prompt, opts = {}) {
+	const rules = opts.rules !== undefined ? opts.rules : loadRules();
+	if (!rules?.skills) return "";
+	const skillsDir = opts.skillsDir ?? path.join(__dirname, "..", "skills");
+
+	const {
+		minConfidenceScore = 3,
+		maxSkillsToShow = 5,
+		showMatchReasons = true,
+	} = rules.config ?? {};
+
+	// Cap the text before the pattern scans (F09), and coerce non-strings (F21).
+	const cappedPrompt = String(prompt).slice(0, MAX_PROMPT_LENGTH);
+	const promptLower = cappedPrompt.toLowerCase();
+	const filePaths = extractFilePaths(cappedPrompt);
+	const skills = rules.skills;
 
 	const matches = [];
 	for (const [name, skill] of Object.entries(skills)) {
-		const match = evaluateSkill(name, skill, prompt, promptLower, filePaths, rules);
-		if (match && match.score >= config.minConfidenceScore) {
+		const match = evaluateSkill(name, skill, cappedPrompt, promptLower, filePaths, rules);
+		if (match && match.score >= minConfidenceScore && skillExists(match.name, skillsDir)) {
 			matches.push(match);
 		}
 	}
@@ -284,29 +320,31 @@ function evaluate(prompt) {
 		return b.priority - a.priority;
 	});
 
-	const topMatches = matches.slice(0, config.maxSkillsToShow);
-	const relatedSkills = getRelatedSkills(topMatches, skills);
+	const topMatches = matches.slice(0, maxSkillsToShow);
+	const relatedSkills = getRelatedSkills(topMatches, skills).filter((n) =>
+		skillExists(n, skillsDir),
+	);
 
 	let output = "<user-prompt-submit-hook>\n";
 	output += "SKILL ACTIVATION REQUIRED\n\n";
 
 	if (filePaths.length > 0) {
-		output += `Detected file paths: ${filePaths.join(", ")}\n\n`;
+		output += `Detected file paths: ${filePaths.map(sanitize).join(", ")}\n\n`;
 	}
 
 	output += "Matched skills (ranked by relevance):\n";
 
 	for (let i = 0; i < topMatches.length; i++) {
 		const match = topMatches[i];
-		const confidence = formatConfidence(match.score, config.minConfidenceScore);
-		output += `${i + 1}. ${match.name} (${confidence} confidence)\n`;
-		if (config.showMatchReasons && match.reasons.length > 0) {
-			output += `   Matched: ${match.reasons.slice(0, 3).join(", ")}\n`;
+		const confidence = formatConfidence(match.score, minConfidenceScore);
+		output += `${i + 1}. ${sanitize(match.name)} (${confidence} confidence)\n`;
+		if (showMatchReasons && match.reasons.length > 0) {
+			output += `   Matched: ${match.reasons.slice(0, 3).map(sanitize).join(", ")}\n`;
 		}
 	}
 
 	if (relatedSkills.length > 0) {
-		output += `\nRelated skills to consider: ${relatedSkills.join(", ")}\n`;
+		output += `\nRelated skills to consider: ${relatedSkills.map(sanitize).join(", ")}\n`;
 	}
 
 	output += "\nBefore implementing, you MUST:\n";
@@ -314,9 +352,9 @@ function evaluate(prompt) {
 	output += "2. ACTIVATE: Invoke the Skill tool for each YES skill\n";
 	output += "3. IMPLEMENT: Only proceed after skill activation\n";
 	output += "\nExample evaluation:\n";
-	output += `- ${topMatches[0].name}: YES - [your reasoning]\n`;
+	output += `- ${sanitize(topMatches[0].name)}: YES - [your reasoning]\n`;
 	if (topMatches.length > 1) {
-		output += `- ${topMatches[1].name}: NO - [your reasoning]\n`;
+		output += `- ${sanitize(topMatches[1].name)}: NO - [your reasoning]\n`;
 	}
 	output += "\nDO NOT skip this step. Invoke relevant skills NOW.\n";
 	output += "</user-prompt-submit-hook>";
@@ -334,18 +372,25 @@ function main() {
 		input += chunk;
 	});
 
+	// F36: never leave stdin errors unhandled (would otherwise hang to the hook timeout).
+	process.stdin.on("error", () => {
+		process.exitCode = 0;
+	});
+
 	process.stdin.on("end", () => {
 		let prompt = "";
 
 		try {
 			const data = JSON.parse(input);
-			prompt = data.prompt || "";
+			// F21: only accept a string prompt; anything else is treated as empty.
+			prompt = typeof data.prompt === "string" ? data.prompt : "";
 		} catch {
 			prompt = input;
 		}
 
-		if (!prompt.trim()) {
-			process.exit(0);
+		if (typeof prompt !== "string" || !prompt.trim()) {
+			process.exitCode = 0;
+			return;
 		}
 
 		try {
@@ -357,8 +402,25 @@ function main() {
 			console.error(`Skill evaluation failed: ${error.message}`);
 		}
 
-		process.exit(0);
+		// F35: set exitCode and let the process drain stdout naturally rather than
+		// process.exit(0), which can truncate a large console.log on POSIX pipes.
+		process.exitCode = 0;
 	});
 }
 
-main();
+if (require.main === module) {
+	main();
+}
+
+module.exports = {
+	matchesGlob,
+	matchesPattern,
+	extractFilePaths,
+	matchDirectoryMapping,
+	evaluateSkill,
+	getRelatedSkills,
+	formatConfidence,
+	skillExists,
+	sanitize,
+	evaluate,
+};