Automated audit: This PR was generated by NLPM, a natural language programming linter, running via claude-code-action. Please evaluate the diff on its merits.
About this audit
This issue summarizes the results of an automated NL-quality scan of this repository by NLPM (Natural Language Programming Manager), a linter for Claude Code artifacts (agents, commands, skills, hooks).
NLPM scores repositories on a 100-point scale — this repo scored 77/100, which is respectable for a hooks-focused demo repo. The 5 PRs below address mechanical bugs only; quality suggestions are listed at the bottom for informational purposes.
No Critical or High security findings were found. The Medium/Low security items are minor and confined to the project's demo nature.
Bugs fixed by PRs
These are mechanical defects — missing fields that Claude Code requires to function correctly. The PRs make the smallest possible change to fix each one.
| # |
Priority |
File |
Issue |
PR |
| 1 |
High |
.claude/agents/llm-ai-agents-and-eng-research.md |
WebSearch used in step 2 instructions but not declared in tools — WebSearch calls fail silently at runtime |
#46 |
| 2 |
High |
.claude/commands/cook.md |
No frontmatter — command is undiscoverable in slash-command UI |
#47 |
| 3 |
High |
.claude/commands/cook_research_only.md |
No frontmatter — command is undiscoverable in slash-command UI |
#47 |
| 4 |
High |
.claude/commands/all_tools.md |
No frontmatter — command is undiscoverable in slash-command UI |
#48 |
| 5 |
High |
.claude/commands/update_status_line.md |
No frontmatter — command is undiscoverable in slash-command UI |
#48 |
| 6-12 |
High |
.claude/commands/agent_prompts/*.md (7 files) |
No frontmatter on any of the 7 agent prompt command files — undiscoverable |
#41 |
| 13 |
Medium |
.claude/commands/sentient.md |
Description "Manage, organize and ships your codebase" does not match demo content that runs rm -rf; risks misrouting in auto-dispatch |
#43 |
Remaining bugs (not covered by PRs — max 5 PRs per run)
The 5 PRs above cover the most impactful bugs. The remaining items from the audit are informational:
prime_tts.md: description appears to already be present in current HEAD — the audit may have been run against an older commit- Security (Medium):
setup.py runtime package install runs on --install-deps flag — consider gating behind explicit user confirmation
- Security (Low):
apps/task-manager/package.json uses ^ semver ranges — consider pinning to exact versions or committing a lockfile
Quality observations (informational, no PRs)
These are improvements that would raise the score but are not bugs:
- 13 thin crypto agents delegate entirely via
Read and Execute to prompt files — the prompt files lack example invocations and output format specs (adds examples would improve score)
CLAUDE.md is effectively empty — given the depth of hooks and agents here, a project context section would help Claude bootstrap faster in new sessions- Several agent files (
work-completion-summary.md, hello-world-agent.md, llm-ai-agents-and-eng-research.md) omit a model declaration
This is a high-quality, well-structured repo. The hooks architecture is genuinely impressive and the crypto agent tier design (haiku/sonnet/opus variants) is a clean pattern. Thank you for sharing it publicly!
Please feel free to close this issue and the PRs if the changes don't fit the project's style or intent — no obligation.
About this audit
This issue summarizes the results of an automated NL-quality scan of this repository by NLPM (Natural Language Programming Manager), a linter for Claude Code artifacts (agents, commands, skills, hooks).
NLPM scores repositories on a 100-point scale — this repo scored 77/100, which is respectable for a hooks-focused demo repo. The 5 PRs below address mechanical bugs only; quality suggestions are listed at the bottom for informational purposes.
No Critical or High security findings were found. The Medium/Low security items are minor and confined to the project's demo nature.
Bugs fixed by PRs
These are mechanical defects — missing fields that Claude Code requires to function correctly. The PRs make the smallest possible change to fix each one.
.claude/agents/llm-ai-agents-and-eng-research.mdWebSearchused in step 2 instructions but not declared intools— WebSearch calls fail silently at runtime.claude/commands/cook.md.claude/commands/cook_research_only.md.claude/commands/all_tools.md.claude/commands/update_status_line.md.claude/commands/agent_prompts/*.md(7 files).claude/commands/sentient.mdrm -rf; risks misrouting in auto-dispatchRemaining bugs (not covered by PRs — max 5 PRs per run)
The 5 PRs above cover the most impactful bugs. The remaining items from the audit are informational:
prime_tts.md: description appears to already be present in current HEAD — the audit may have been run against an older commitsetup.pyruntime package install runs on--install-depsflag — consider gating behind explicit user confirmationapps/task-manager/package.jsonuses^semver ranges — consider pinning to exact versions or committing a lockfileQuality observations (informational, no PRs)
These are improvements that would raise the score but are not bugs:
Read and Executeto prompt files — the prompt files lack example invocations and output format specs (adds examples would improve score)CLAUDE.mdis effectively empty — given the depth of hooks and agents here, a project context section would help Claude bootstrap faster in new sessionswork-completion-summary.md,hello-world-agent.md,llm-ai-agents-and-eng-research.md) omit amodeldeclarationThis is a high-quality, well-structured repo. The hooks architecture is genuinely impressive and the crypto agent tier design (haiku/sonnet/opus variants) is a clean pattern. Thank you for sharing it publicly!
Please feel free to close this issue and the PRs if the changes don't fit the project's style or intent — no obligation.