📦
Audit History
skill-creator - 4 audits
Version comparison
Capability and finding changes across audited versions, newest first.
Audit version 4 Latest
Jul 7, 2026, 02:54 AM
Most static command and network alerts are false positives from Markdown examples, placeholder URLs, or benign local validation code. Confirmed issues center on global symlink installation into hidden AI agent directories, missing skill-name path validation, and packaging behavior that may include unintended files.
8
Files scanned
1,688
Lines analyzed
26
Review items
0
False positives ignored
Confirmed security concerns (3)
High
Unsanitized Skill Name Can Escape Target Directory
The initializer combines the requested path and skill name before creating directories and files, but it does not validate the skill name. A name containing path separators or an absolute path could write the generated skill outside the intended output directory.
The code and documented shell workflow derive filesystem paths from user-controlled skill names without enforcing the stated hyphen-case rules. The resulting operations create directories and write files.
Medium
Misleading Safe Risk Metadata
The skill frontmatter declares risk as safe while the workflow supports global installation through symlinks in hidden agent directories. That mismatch can understate the review needed before installing a community skill.
The safe label is explicit in frontmatter, and later sections instruct global symlink installation into per-user agent configuration paths. The behavior is intentional but security-sensitive.
Medium
Package Archive May Include Symlink Targets
The packaging script recursively writes every file found under the skill folder into an archive. It does not reject symlinks or filter unexpected files, which can leak outside file contents if a skill folder contains symlinked files.
Path.rglob combined with is_file can include symlinked files, and zipfile.write has no visible filter here. The risk depends on packaging a folder that already contains unsafe links or unrelated files.
Capability review items (20)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
High
Hidden file in home directory
ln -sf "$(pwd)/.github/skills/skill-creator" ~/.copilot/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Hidden file in home directory
ln -sf "$(pwd)/.claude/skills/skill-creator" ~/.claude/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$(pwd)/.github/skills/skill-creator" ~/.copilot/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$(pwd)/.claude/skills/skill-creator" ~/.claude/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Hidden file in home directory
- [ ] **Global installation** - Create symlinks in `~/.copilot/skills/` (works everywhere)
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Hidden file in home directory
- Always use absolute paths: `ln -sf /full/path/to/source ~/.copilot/skills/name`
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$SKILLS_REPO/.github/skills/$SKILL_NAME" \
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$SKILLS_REPO/.claude/skills/$SKILL_NAME" \
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$SKILLS_REPO/.codex/skills/$SKILL_NAME" \
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
- Always use absolute paths: `ln -sf /full/path/to/source ~/.copilot/skills/name`
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
Medium
Hidden file access
ln -sf "$(pwd)/.github/skills/skill-creator" ~/.copilot/skills/skill-creator
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$(pwd)/.claude/skills/skill-creator" ~/.claude/skills/skill-creator
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
- [ ] **Global installation** - Create symlinks in `~/.copilot/skills/` (works everywhere)
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$SKILLS_REPO/.github/skills/$SKILL_NAME" \
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
"$HOME/.copilot/skills/$SKILL_NAME"
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$SKILLS_REPO/.claude/skills/$SKILL_NAME" \
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
"$HOME/.claude/skills/$SKILL_NAME"
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$SKILLS_REPO/.codex/skills/$SKILL_NAME" \
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
"$HOME/.codex/skills/$SKILL_NAME"
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
- Always use absolute paths: `ln -sf /full/path/to/source ~/.copilot/skills/name`
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Risk Factors
⚙️ External commands (98)
README.md:45 README.md:48 scripts/init_skill.py:73 scripts/init_skill.py:74 scripts/init_skill.py:84 SKILL.md:42-73 SKILL.md:73-87 SKILL.md:87-89 SKILL.md:89-98 SKILL.md:98-100 SKILL.md:100-105 SKILL.md:105-107 SKILL.md:107-110 SKILL.md:110-122 SKILL.md:122-152 SKILL.md:152-154 SKILL.md:154-157 SKILL.md:157-164 SKILL.md:164-183 SKILL.md:183-185 SKILL.md:185-188 SKILL.md:188-196 SKILL.md:196-200 SKILL.md:200-216 SKILL.md:216-221 SKILL.md:221 SKILL.md:223-224 SKILL.md:224-225 SKILL.md:225-226 SKILL.md:226-227 SKILL.md:227-228 SKILL.md:228-229 SKILL.md:229-231 SKILL.md:231-237 SKILL.md:237-242 SKILL.md:242-269 SKILL.md:269-272 SKILL.md:272-282 SKILL.md:282-287 SKILL.md:287-289 SKILL.md:289-292 SKILL.md:292-299 SKILL.md:299-303 SKILL.md:303-309 SKILL.md:309-312 SKILL.md:312-319 SKILL.md:319-334 SKILL.md:334-336 SKILL.md:336-339 SKILL.md:339-346 SKILL.md:346-351 SKILL.md:351-352 SKILL.md:352-358 SKILL.md:358-377 SKILL.md:377-381 SKILL.md:381-402 SKILL.md:402-406 SKILL.md:406-411 SKILL.md:411-416 SKILL.md:416-418 SKILL.md:418-421 SKILL.md:421-428 SKILL.md:428-432 SKILL.md:432-459 SKILL.md:459-466 SKILL.md:466-473 SKILL.md:473-478 SKILL.md:478-487 SKILL.md:487-492 SKILL.md:492-503 SKILL.md:503-508 SKILL.md:508-516 SKILL.md:516-525 SKILL.md:525-526 SKILL.md:526-527 SKILL.md:527-528 SKILL.md:528-533 SKILL.md:533-534 SKILL.md:534-535 SKILL.md:535-540 SKILL.md:540-541 SKILL.md:541-542 SKILL.md:542-547 SKILL.md:547-552 SKILL.md:552-562 SKILL.md:562-563 SKILL.md:563-564 SKILL.md:564-592 SKILL.md:592-593 SKILL.md:61 SKILL.md:71 SKILL.md:72 SKILL.md:202 SKILL.md:247 SKILL.md:261 SKILL.md:42-73 SKILL.md:200-216 SKILL.md:242-269
🌐 Network access (4)
📁 Filesystem access (38)
README.md:45 README.md:48 README.md:45 README.md:48 README.md:45 README.md:48 scripts/package_skill.py:14 scripts/package_skill.py:68 SKILL.md:352 SKILL.md:408 SKILL.md:409 SKILL.md:410 SKILL.md:509 SKILL.md:552 SKILL.md:52 SKILL.md:56 SKILL.md:65 SKILL.md:352 SKILL.md:384 SKILL.md:385 SKILL.md:391 SKILL.md:392 SKILL.md:398 SKILL.md:399 SKILL.md:408 SKILL.md:409 SKILL.md:410 SKILL.md:509 SKILL.md:552 SKILL.md:48 SKILL.md:61 SKILL.md:408 SKILL.md:409 SKILL.md:410 SKILL.md:384 SKILL.md:391 SKILL.md:398 SKILL.md:552
Audited by: codex
Jul 7, 2026, 02:54 AM
Most static command and network alerts are false positives from Markdown examples, placeholder URLs, or benign local validation code. Confirmed issues center on global symlink installation into hidden AI agent directories, missing skill-name path validation, and packaging behavior that may include unintended files.
8
Files scanned
1,688
Lines analyzed
26
Review items
0
False positives ignored
Confirmed security concerns (3)
High
Unsanitized Skill Name Can Escape Target Directory
The initializer combines the requested path and skill name before creating directories and files, but it does not validate the skill name. A name containing path separators or an absolute path could write the generated skill outside the intended output directory.
The code and documented shell workflow derive filesystem paths from user-controlled skill names without enforcing the stated hyphen-case rules. The resulting operations create directories and write files.
Medium
Misleading Safe Risk Metadata
The skill frontmatter declares risk as safe while the workflow supports global installation through symlinks in hidden agent directories. That mismatch can understate the review needed before installing a community skill.
The safe label is explicit in frontmatter, and later sections instruct global symlink installation into per-user agent configuration paths. The behavior is intentional but security-sensitive.
Medium
Package Archive May Include Symlink Targets
The packaging script recursively writes every file found under the skill folder into an archive. It does not reject symlinks or filter unexpected files, which can leak outside file contents if a skill folder contains symlinked files.
Path.rglob combined with is_file can include symlinked files, and zipfile.write has no visible filter here. The risk depends on packaging a folder that already contains unsafe links or unrelated files.
Capability review items (20)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
High
Hidden file in home directory
ln -sf "$(pwd)/.github/skills/skill-creator" ~/.copilot/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Hidden file in home directory
ln -sf "$(pwd)/.claude/skills/skill-creator" ~/.claude/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$(pwd)/.github/skills/skill-creator" ~/.copilot/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$(pwd)/.claude/skills/skill-creator" ~/.claude/skills/skill-creator
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Hidden file in home directory
- [ ] **Global installation** - Create symlinks in `~/.copilot/skills/` (works everywhere)
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Hidden file in home directory
- Always use absolute paths: `ln -sf /full/path/to/source ~/.copilot/skills/name`
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$SKILLS_REPO/.github/skills/$SKILL_NAME" \
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$SKILLS_REPO/.claude/skills/$SKILL_NAME" \
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
ln -sf "$SKILLS_REPO/.codex/skills/$SKILL_NAME" \
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
High
Symlink creation
- Always use absolute paths: `ln -sf /full/path/to/source ~/.copilot/skills/name`
This global installation path writes a symlink into hidden AI tool skill directories under the user home directory. If run, it can replace an existing skill link and persist agent behavior across projects.
Medium
Hidden file access
ln -sf "$(pwd)/.github/skills/skill-creator" ~/.copilot/skills/skill-creator
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$(pwd)/.claude/skills/skill-creator" ~/.claude/skills/skill-creator
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
- [ ] **Global installation** - Create symlinks in `~/.copilot/skills/` (works everywhere)
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$SKILLS_REPO/.github/skills/$SKILL_NAME" \
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
"$HOME/.copilot/skills/$SKILL_NAME"
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$SKILLS_REPO/.claude/skills/$SKILL_NAME" \
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
"$HOME/.claude/skills/$SKILL_NAME"
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
ln -sf "$SKILLS_REPO/.codex/skills/$SKILL_NAME" \
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
"$HOME/.codex/skills/$SKILL_NAME"
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Medium
Hidden file access
- Always use absolute paths: `ln -sf /full/path/to/source ~/.copilot/skills/name`
This finding is part of the documented global installation flow that writes into hidden agent skill directories. The operation is user-directed, but it still modifies sensitive per-user agent configuration.
Risk Factors
⚙️ External commands (98)
README.md:45 README.md:48 scripts/init_skill.py:73 scripts/init_skill.py:74 scripts/init_skill.py:84 SKILL.md:42-73 SKILL.md:73-87 SKILL.md:87-89 SKILL.md:89-98 SKILL.md:98-100 SKILL.md:100-105 SKILL.md:105-107 SKILL.md:107-110 SKILL.md:110-122 SKILL.md:122-152 SKILL.md:152-154 SKILL.md:154-157 SKILL.md:157-164 SKILL.md:164-183 SKILL.md:183-185 SKILL.md:185-188 SKILL.md:188-196 SKILL.md:196-200 SKILL.md:200-216 SKILL.md:216-221 SKILL.md:221 SKILL.md:223-224 SKILL.md:224-225 SKILL.md:225-226 SKILL.md:226-227 SKILL.md:227-228 SKILL.md:228-229 SKILL.md:229-231 SKILL.md:231-237 SKILL.md:237-242 SKILL.md:242-269 SKILL.md:269-272 SKILL.md:272-282 SKILL.md:282-287 SKILL.md:287-289 SKILL.md:289-292 SKILL.md:292-299 SKILL.md:299-303 SKILL.md:303-309 SKILL.md:309-312 SKILL.md:312-319 SKILL.md:319-334 SKILL.md:334-336 SKILL.md:336-339 SKILL.md:339-346 SKILL.md:346-351 SKILL.md:351-352 SKILL.md:352-358 SKILL.md:358-377 SKILL.md:377-381 SKILL.md:381-402 SKILL.md:402-406 SKILL.md:406-411 SKILL.md:411-416 SKILL.md:416-418 SKILL.md:418-421 SKILL.md:421-428 SKILL.md:428-432 SKILL.md:432-459 SKILL.md:459-466 SKILL.md:466-473 SKILL.md:473-478 SKILL.md:478-487 SKILL.md:487-492 SKILL.md:492-503 SKILL.md:503-508 SKILL.md:508-516 SKILL.md:516-525 SKILL.md:525-526 SKILL.md:526-527 SKILL.md:527-528 SKILL.md:528-533 SKILL.md:533-534 SKILL.md:534-535 SKILL.md:535-540 SKILL.md:540-541 SKILL.md:541-542 SKILL.md:542-547 SKILL.md:547-552 SKILL.md:552-562 SKILL.md:562-563 SKILL.md:563-564 SKILL.md:564-592 SKILL.md:592-593 SKILL.md:61 SKILL.md:71 SKILL.md:72 SKILL.md:202 SKILL.md:247 SKILL.md:261 SKILL.md:42-73 SKILL.md:200-216 SKILL.md:242-269
🌐 Network access (4)
📁 Filesystem access (38)
README.md:45 README.md:48 README.md:45 README.md:48 README.md:45 README.md:48 scripts/package_skill.py:14 scripts/package_skill.py:68 SKILL.md:352 SKILL.md:408 SKILL.md:409 SKILL.md:410 SKILL.md:509 SKILL.md:552 SKILL.md:52 SKILL.md:56 SKILL.md:65 SKILL.md:352 SKILL.md:384 SKILL.md:385 SKILL.md:391 SKILL.md:392 SKILL.md:398 SKILL.md:399 SKILL.md:408 SKILL.md:409 SKILL.md:410 SKILL.md:509 SKILL.md:552 SKILL.md:48 SKILL.md:61 SKILL.md:408 SKILL.md:409 SKILL.md:410 SKILL.md:384 SKILL.md:391 SKILL.md:398 SKILL.md:552
Audited by: codex
Jun 30, 2026, 05:50 PM
Static analysis reported many command, network, and filesystem patterns, but most command and network hits are markdown examples, license URLs, or template text. AI review found no evidence of malicious network exfiltration or prompt injection, but the shipped scripts and installation guidance can write files, create archives, and create global symlinks, so publication should include a filesystem warning.
8
Files scanned
1,688
Lines analyzed
7
Review items
3
False positives ignored
Confirmed security concerns (3)
Medium
Path-Controlled Skill Scaffolding Writes Files
The initializer combines a user-provided path and skill name, creates the resulting directory, writes SKILL.md, creates resource directories, writes placeholder files, and marks an example script executable. This is expected for a scaffolding tool, but it can modify arbitrary user-writable filesystem locations if arguments are careless or adversarial.
The script directly creates directories and writes files based on command-line arguments. The behavior is legitimate scaffolding, but the filesystem impact is real and not limited to a fixed safe directory.
Medium
Global Symlink Installation Guidance Can Overwrite Skill Links
The documentation recommends forced symlink creation into assistant home directories. This is a legitimate installation pattern, but ln -sf can replace existing symlinks and affects global assistant behavior.
The cited lines show forced symlink commands and verification of home-directory skill paths. The commands are presented as installation steps, so the risk is contextual rather than malicious.
Medium
Archive Packaging Can Include Unintended Files
The packager walks every file under the selected skill directory and writes it into a .skill archive after basic validation. This can accidentally include secrets or unrelated files if the source directory is not reviewed first.
The recursive archive creation is explicit and does not filter sensitive filenames. There is no upload path, so the concern is accidental disclosure through later sharing of the archive.
Static false positives ignored (3)
These static matches were dismissed by semantic review or matched schema-only tokens, so they are shown for transparency but do not drive the quality score.
Low
Static Network Findings Are Documentation Links
Hardcoded URL findings occur in the license, clone examples, and references. No executable HTTP client, fetch call, or credential-bearing outbound request was found in the inspected files.
The evidence is limited to URLs in text and example clone commands. A targeted search found no requests, urllib, curl, wget, socket, fetch, or similar executable network client in the scripts.
Low
Static Weak Cryptography Findings Are Text Matches
The weak-cryptography alerts appear to be triggered by ordinary words inside license text, documentation, validation messages, and template guidance. No cryptographic API use was found in the reviewed scripts.
The inspected code imports pathlib, zipfile, yaml, re, os, and sys, but not cryptographic libraries. The flagged lines are not cryptographic operations.
Low
Markdown Command Examples Were Counted As Execution
Many external-command findings are shell examples or illustrative markdown blocks. They are not executed by the package scripts, although users should still review commands before running them.
The cited command text is contained in markdown documentation. The Python scripts do not call subprocess, os.system, eval, or exec.
Risk Factors
⚡ Contains scripts (3)
📁 Filesystem access (5)
⚙️ External commands (3)
🌐 Network access (3)
Detected Patterns
Filesystem Writes From User-Controlled ArgumentsForced Symlink Creation In Home Directories
Audited by: codex
Feb 25, 2026, 02:55 AM
Static scanner flagged 274 patterns but evaluation reveals these are false positives. Markdown code blocks trigger 'shell execution' alerts, 'MD' substrings trigger 'weak crypto' alerts, and documentation URLs trigger 'network' alerts. Actual Python scripts are legitimate scaffolding tools with no credential access or command injection vulnerabilities.
8
Files scanned
1,689
Lines analyzed
6
Review items
0
False positives ignored
Confirmed security concerns (2)
Low
Documentation References Flagged as Network Calls
Static scanner flagged hardcoded URLs in LICENSE.txt and README.md. These are documentation references, not actual network calls in code.
URLs are plain text in documentation files, not network calls in executable code
Low
Substring Matching Error for Weak Crypto
Scanner incorrectly flags 'MD' substring in words like 'Provided', 'Derivative', 'designated' as 'MD5' weak cryptographic algorithm.
Pattern matching error - no weak crypto algorithms present in code
Capability review items (2)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
Low
Markdown Code Blocks Flagged as Shell Execution
Backticks in markdown files (README.md, SKILL.md) are for code formatting, not actual shell execution. Backticks in Python string literals are documentation examples.
Verified by reading code - no subprocess calls found, only markdown code blocks and string literals
Low
Documentation About File Operations Flagged
References to ~/.claude, ~/.copilot directories and symlink creation are legitimate skill installation instructions, not malicious file operations.
Symlink creation is documented best practice for skill installation
Risk Factors
📁 Filesystem access (4)
⚡ Contains scripts (3)
Audited by: claude