Convert selected books into structured Agent Skill packs with atomic skills, combo skills, routers, references, and validation workflows.
SaferSkills independently audited book-skills-creator (Agent Skill) and scored it 82/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 2 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 2 flagged
The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
A book is not one capability. Discover reusable procedures and workflows in the user's explicitly selected sources, then build the smallest complete, non-redundant skill pack.
python scripts/book_skills.py inspect <selected paths> --json. Confirm limits and skipped inputs. Optional local tools are reported by preflight.python scripts/book_skills.py extract <selected paths> --output .book_skills_work --json. Read .book_skills_work/sources.json locally. It contains ordered evidence units and source text; protect it as sensitive. metadata.json and full_text.txt are compatibility/debug artifacts.python scripts/book_skills.py plan .book_skills_work/sources.json <pack-name> --output .book_skills_work/plan.json. Fill the draft plan's candidates, atomic, combo, routes, references, overlaps, conflicts, uncertainties, validation_risks, and behavior_tests from source evidence; inspect its extraction_gaps. Use the schema in docs/SPECIFICATION.md and the sample examples/sample-pack/pack.json. Record rejected candidates and reasons. Resolve material ambiguity with the user when needed. Set status to ready only after a deliberate architecture exists.python scripts/book_skills.py build .book_skills_work/plan.json --output <pack-directory> --json. The renderer writes exactly the units in the plan; it inserts no example skills. pack.json becomes the pack's canonical structure.python scripts/book_skills.py validate <pack-directory> --extraction .book_skills_work/sources.json --json. Fix errors. Review warnings about overlap, unresolved conflict, copied passages, and grounding. Evaluate the manifest's positive, negative, combo, reference, ambiguous, and unsupported routing cases with an agent or human; the CLI checks their structure, not their semantics.Deterministic helpers enforce selection, extraction limits, IDs, serialization, graph integrity, file consistency, and referenced evidence. The agent must judge meaning, decomposition, grounding quality, conflicts, and behavioral usefulness. Passing structural validation alone does not prove semantic fidelity.
For source changes, extract the same selected paths again and run book-skills update <pack-directory> <fresh-sources.json> --json. Review affected units, revise the plan, rebuild, and validate. No semantic regeneration occurs automatically.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.