What exactly is broken in popular skills: references to files that are not bundled, actions taken without human approval, processes with no stopping condition. Everything here is measured from the skill text rather than judged by a model, so every line can be checked.
GeneratorGitHubwill not run
Generate wiki docs + Mermaid diagrams for any codebase.
- References files that are not bundled: <link>, architecture.md, modules/<module>.md
- The response is described with custom markup (9 tags): a typed call is more reliable
- 5 referenced file(s) missing: <link>, architecture.md, modules/<module>.md
NousResearch/hermes-agent · ★ 245 156
Procedure
Fast BPE/WordPiece tokenization and custom vocab training.
- 11 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
Integration
HuggingFace hf CLI: search/download/upload models, datasets.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHubCloudflare
Versioned site deploys to GitHub/Cloudflare/Netlify Pages.
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- The skill tells the model to perform an irreversible action with no human approval
- The response is described with custom markup (10 tags): a typed call is more reliable
NousResearch/hermes-agent · ★ 245 156
Integration
Configure and troubleshoot Honcho memory for Hermes.
- 10 top-level sections: this looks like several domains in one skill
- The response is described with custom markup (6 tags): a typed call is more reliable
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
Procedurewill not run
Debug Python: pdb REPL + debugpy remote (DAP).
- References files that are not bundled: scripts/run_tests.sh
- 12 top-level sections: this looks like several domains in one skill
- 1 referenced file(s) missing: scripts/run_tests.sh
NousResearch/hermes-agent · ★ 245 156
Integration
Build integrated IS/BS/CF financial workbooks in Excel.
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 12 top-level sections: this looks like several domains in one skill
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
Generatorwill not run
Create meme PNGs from templates with Pillow text overlay.
- References files that are not bundled: scripts/.cache/
- 1 referenced file(s) missing: scripts/.cache/
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
IntegrationStripe
Agent payments via Stripe Link — cards, SPT, approvals.
- The skill tells the model to perform an irreversible action with no human approval
- No condition that starts the skill
- 3 mutating operations with no state check
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHub
Clean training loops with built-in distributed support.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
AnalyzerExcelwill not run
Build comparable-company valuation workbooks in Excel.
- References files that are not bundled: examples/comps_example.xlsx
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 22 top-level sections: this looks like several domains in one skill
NousResearch/hermes-agent · ★ 245 156
Procedure
Provision Twilio numbers, SMS/MMS, and AI outbound calls.
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 14 top-level sections: this looks like several domains in one skill
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
GeneratorGitHub
Speed up long-sequence transformer training and inference.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHub
Passive recon of subdomains, SSL certs, WHOIS, and DNS.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
Procedurewill not run
Knowledge comics (知识漫画): educational, biography, tutorial.
- References files that are not bundled: references/art-styles/<style>.md, references/tones/<tone>.md, references/presets/<preset>.md
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- The response is described with custom markup (6 tags): a typed call is more reliable
NousResearch/hermes-agent · ★ 245 156
Generator
Hand-drawn Excalidraw JSON diagrams (arch, flow, seq).
- Does not say what the result is
- Does not say what the process needs to start
NousResearch/hermes-agent · ★ 245 156
Reference
Frontend design guidance, upstream-maintained (impeccable).
- Prose only: no discrete steps
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
IntegrationGitHubDocker
High-throughput LLM inference on NVIDIA GPUs.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
GeneratorExcelwill not run
Build leveraged buyout workbooks with IRR/MOIC in Excel.
- References files that are not bundled: examples/LBO_Model.xlsx
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 11 top-level sections: this looks like several domains in one skill
NousResearch/hermes-agent · ★ 245 156
Procedure
On-demand GPU cloud instances for ML training.
- 17 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
IntegrationShopifywill not run
Shop catalog search, checkout, order tracking, returns.
- References files that are not bundled: url
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 12 top-level sections: this looks like several domains in one skill
NousResearch/hermes-agent · ★ 245 156
Procedurewill not run
Infographics: 21 layouts x 21 styles (信息图, 可视化).
- References files that are not bundled: references/layouts/<layout>.md, references/styles/<style>.md
- 11 top-level sections: this looks like several domains in one skill
- The response is described with custom markup (8 tags): a typed call is more reliable
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHub
Run PyTorch training across GPUs with minimal changes.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHub
Pretrain LLMs at scale with PyTorch 4D parallelism.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHub
Poll RSS, JSON APIs, and GitHub with watermark dedup.
- The skill tells the model to perform an irreversible action with no human approval
- The response is described with custom markup (5 tags): a typed call is more reliable
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156