OpenSpace dir-verify-parallel-shell-audit
Diagnose and resolve file access issues by verifying directory structure, reading multiple files in parallel batches, running combined shell property checks (head/wc-l) in the same iteration as directory listing, and validating TypeScript import patterns — all with built-in guidance to minimize total iteration count.
git clone https://github.com/HKUDS/OpenSpace
T=$(mktemp -d) && git clone --depth=1 https://github.com/HKUDS/OpenSpace "$T" && mkdir -p ~/.claude/skills && cp -r "$T/showcase/skills/dir-verify-parallel-shell-audit" ~/.claude/skills/hkuds-openspace-dir-verify-parallel-shell-audit && rm -rf "$T"
showcase/skills/dir-verify-parallel-shell-audit/SKILL.mdVerify Directory Structure — Parallel Audit with Shell Property Checks
This skill provides a systematic approach to diagnosing file access issues and auditing source trees efficiently. It extends the parallel batch reading approach with an integrated shell property check phase that combines
head -N and wc -l into a single run_shell call, co-scheduled with directory listing in Phase 1. This eliminates a common extra iteration that would otherwise be required to check file properties such as line counts and leading content (e.g., CSS/config file headers).
Steps
Phase 1 — Directory Verification + Shell Property Checks (Combined)
Key optimization: Fire
calls andlist_dirproperty checks in the same iteration. This avoids a dedicated iteration later just to learn "how many lines does this file have?" or "what is the first line of this CSS file?".run_shell
-
Identify Target Paths: Collect every file or directory path involved in the operation. Prefer absolute paths to eliminate ambiguity. At this stage, also note any files whose properties (line count, header content) you already know you will need — typically CSS, config, or data files.
-
Issue Parallel
+list_dir
Calls Simultaneously:run_shell- Call
on each relevant directory.list_dir - In the same iteration, issue one or more
calls that bundlerun_shell
andhead
for known target files using a single compound shell command.wc -l
Iteration 1 (parallel): list_dir(src/components) list_dir(src/styles) run_shell("head -5 src/styles/main.css && echo '---' && wc -l src/styles/main.css") run_shell("head -3 vite.config.ts && echo '---' && wc -l src/main.ts")Compound shell pattern (combine head + wc + any other checks for a file):
# Single run_shell call — multiple files, multiple checks, zero extra iterations head -10 path/to/file.css && echo "LINE_COUNT:" && wc -l path/to/file.css && \ echo "---" && head -5 path/to/config.ts && echo "LINE_COUNT:" && wc -l path/to/config.ts - Call
-
Verify Existence & Permissions: Confirm directories exist and that the current user has read (and, if needed, write/execute) permissions. If a directory is missing, create it before proceeding (
). If permissions are insufficient, adjust them before retrying.mkdir -p -
Build a File Manifest: From the
output, compile the complete list of files to be audited. Group them logically (e.g., by subdirectory or file type) so they can be read in parallel batches in Phase 2.list_dir -
Record Shell Check Results: Parse and store the
/head
output alongside the manifest. This data is immediately available for the final audit report without any additional iterations.wc -l
Phase 2 — Parallel Batch File Reading
Why this matters: Reading files sequentially costs one iteration per file. Reading them in parallel batches collapses N files into ⌈N/batch_size⌉ iterations — a significant speedup when auditing dozens of files.
-
Determine Batch Size: Default to reading 5–10 files per batch. Reduce the batch size if individual files are very large; increase it for small config/index files. Files whose content was partially previewed via
in Phase 1 should still be fully read here if their complete content is needed.head -
Issue Parallel
Calls: Within each batch, invokeread_file
for all files simultaneously (a single iteration). Do not wait for one file before requesting the next within the same batch.read_fileBatch 1 (parallel): read_file(src/components/A.tsx) read_file(src/components/B.tsx) read_file(src/components/C.tsx) read_file(src/components/D.tsx) read_file(src/components/E.tsx) Batch 2 (parallel): read_file(src/components/F.tsx) ... -
Collect & Triage Results: After each batch completes, scan results for errors (file-not-found, permission denied) and flag them immediately. Do not abort the remaining batches; continue and aggregate all findings.
-
Re-verify Missing Files: For any file reported as missing, re-run
on its parent directory to confirm whether the file truly does not exist or whether the path was wrong. Correct the path and retry the single file if necessary.list_dir
Phase 3 — TypeScript Import Validation
This phase targets a common class of structural bugs found when auditing TypeScript source trees.
-
Identify Import Statements: For every
/.ts
file read in Phase 2, extract all.tsx
lines.import -
Classify Import Kind — Value vs. Type:
Pattern Correct usage import Foo from '...'Runtime value (class, function, object) import { Foo } from '...'Runtime value export import type Foo from '...'Type-only import (erased at compile time) import type { Foo } from '...'Type-only named import import { type Foo } from '...'Inline type modifier (TS 4.5+) -
Flag Value Imports Used Only as Types: If an identifier imported without
is used only in type positions (type
,: Foo
,as Foo
, generic parameters), it should be converted toimplements Foo
. This prevents accidental runtime dependencies and satisfiesimport type
/verbatimModuleSyntax
compiler flags.isolatedModules// ❌ Before — value import used only as a type import { UserProfile } from './types'; const handler = (u: UserProfile) => { ... }; // ✅ After — type-only import import type { UserProfile } from './types'; const handler = (u: UserProfile) => { ... }; -
Flag Duplicate Imports: Detect cases where the same module is imported more than once in a file (often from copy-paste). Merge them into a single import statement.
// ❌ Duplicate import { A } from './utils'; import { B } from './utils'; // ✅ Merged import { A, B } from './utils'; -
Verify Barrel / Index Exports: When a directory contains an
(barrel file), confirm that every component or module expected to be publicly accessible is re-exported. Cross-reference against the file manifest from Phase 1.index.ts -
Apply Fixes & Re-audit: Apply all import corrections as a single write pass per file. After writing, re-read the affected files (parallel batch) to confirm the fixes took effect.
Phase 4 — Final Validation & Report
-
Retry the Original Operation: With the directory structure confirmed and import issues resolved, retry the operation that originally failed.
-
Log an Audit Report: Produce a concise summary covering:
- Total files enumerated and read
- Number of batches used (highlight the iteration savings vs. sequential)
- Shell property check results (line counts, file headers) gathered in Phase 1 at zero extra iteration cost
- Import issues found and fixed (value→type conversions, duplicate merges, missing barrel exports)
- Any remaining unresolved issues with recommended next steps
Quick Reference: Iteration Collapse Patterns
# Sequential (slow) — N+K iterations for N files + K property checks list_dir(dir) → wait run_shell(wc -l ...) → wait ← extra iteration that can be eliminated read_file(file1) → wait read_file(file2) → wait ... # Optimized (fast) — 1 + ⌈N/B⌉ iterations [list_dir(dir), run_shell("head -5 f1 && wc -l f1 && head -5 f2 && wc -l f2")] ← 1 iteration (Phase 1) [read_file(file1), read_file(file2), ..., read_file(fileB)] ← 1 iteration (Phase 2, batch 1) [read_file(fileB+1), ...] ← 1 iteration (Phase 2, batch 2)
Compound Shell Command Templates
# Check header + line count for a single file head -10 /path/to/file.css && echo "LINES:" && wc -l /path/to/file.css # Check multiple files in one run_shell call head -5 /path/to/a.css && echo "LINES_A:" && wc -l /path/to/a.css && \ echo "===" && \ head -5 /path/to/b.ts && echo "LINES_B:" && wc -l /path/to/b.ts # Verify a file starts with expected content (e.g., CSS theme comment) head -1 /path/to/styles.css | grep -q "Theme Colors" && echo "OK" || echo "MISSING HEADER" # Count files matching a pattern in a directory ls src/components/*.tsx | wc -l
Best Practices
- Co-schedule shell checks with
: Anylist_dir
orwc -l
call you know you'll need belongs in the same iteration as your initialhead -N
calls — never in a dedicated follow-up iteration.list_dir - Bundle multiple shell checks into one
: Userun_shell
or&&
to chain head/wc/grep commands for multiple files into a single;
invocation. Eachrun_shell
call costs one tool-call slot; batching them saves slots.run_shell - Always enumerate before reading:
first, then batch-read. Never guess file names.list_dir - Use absolute paths throughout to avoid working-directory ambiguity.
- Batch by locality: Group files from the same directory in the same batch to keep context coherent.
- Type imports are zero-cost at runtime: Prefer
for any symbol used exclusively as a TypeScript type. This is especially important in projects withimport type
.isolatedModules: true - Barrel files are contracts: Treat
exports as the public API of a module. Missing exports are bugs even if the underlying file is correct.index.ts - Log intermediate findings: If issues persist after Phase 3, dump the full audit log for offline debugging.
Common Fix Categories (Checklist)
- Directory missing → create with
mkdir -p - Permissions insufficient →
/chmod
or escalatechown - Path typo → re-enumerate with
, correct pathlist_dir - File has wrong line count → inspect with
(already gathered in Phase 1 at no extra cost)wc -l - File has wrong header content → inspect with
(already gathered in Phase 1 at no extra cost)head -N - Value import used as type → add
import type - Duplicate imports → merge into single statement
- Missing barrel export → add re-export to
index.ts - Circular import → restructure module boundaries