setup-author-toolkit
Set up or validate a v3.1 author-toolkit book project while preserving manuscript, evidence, and registered paths.
How do I install this agent skill?
npx skills add https://github.com/shengchen1998/author-toolkit --skill setup-author-toolkitIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill is a comprehensive author toolkit designed to manage and validate book projects. It performs file system operations such as project initialization, manuscript tracking, and report generation using local Node.js scripts. No malicious patterns, external downloads, or data exfiltration attempts were detected.
- Socketpass
No alerts
- Snykwarn
Risk: MEDIUM · 1 issue
What does this agent skill do?
Setup Author Toolkit
Toolkit skills from one installation are siblings. For this skill, <setup-skill-root> is its own installed directory. Register one book project and prepare its shared persistent files. Accept only the project root, an optional manuscript path, and optional language or project metadata. There is no entry-stage or workflow argument. Perform the same setup for every later creation or revision strategy. For setup, run node <setup-skill-root>/scripts/terminology.mjs --preset setup-report to read only the rows needed for setup report interpretation and entry routing from Core terms and Contract values in the bundled controlled workflow terminology; do not read the full terminology file. When setup diagnostics require definitions beyond that preset, use node <setup-skill-root>/scripts/terminology.mjs --preset setup for the complete setup selection. Consult its Operation identifiers section by exact identifier with node <setup-skill-root>/scripts/terminology.mjs --section "Operation identifiers" only when a reported pending action, group of unverified approvals, or backup-index difference uses one.
Setup
- Check only whether
<project-root>/.author-toolkit/project.jsonexists. Runnode <setup-skill-root>/scripts/setup-report.mjs begin --jsonand retain itsreport_store; this creates a private temporary report store outside the writing project. After a store is created, runnode <setup-skill-root>/scripts/setup-report.mjs cleanup --report-store <report-store>before any stop prior to rendering, then report the original failure. Runnode <setup-skill-root>/scripts/setup-report.mjs capture --report-store <report-store> --kind doctor -- --require-sibling author-project; capture appends--json, relays the doctor result, and preserves its exit status. If either installation check fails, report an incomplete toolkit installation and stop; this check creates no project files. Defer project-entry installation and verification to step 3. Select exactly one branch after this check: when the manifest exists, read only Existing project; otherwise read only New project. Keep every shared runtime contract in this installed skill'sassets/references/payload and resolve it directly from that location.assets/payload-manifest.jsonrecords the toolkit version and is the authoritative inventory and SHA-256 manifest for this skill's regular payload files, including every CLI's internal modules and licenses. Do not maintain a manual file list, copy shared contracts into the writing project, printpayload-manifest.jsonor the complete manifest, or emit complete session or backup listings. - Follow the selected branch's initialization procedure, then return here for the remaining common steps. Stop for a missing installed file or initialization failure; the existing-project branch explicitly handles exit 3 after an accepted v3.1 schema.
- On every setup, run
node <setup-skill-root>/scripts/setup-report.mjs capture --report-store <report-store> --kind entry-files -- --project-root <project-root>. The capturedinstall-entry-filescommand replaces<project-root>/AGENTS.mdwith a regular file containingassets/AGENTS.md, and replaces any existingCLAUDE.mdfile or symlink with a relative symlink namedCLAUDE.mdpointing toAGENTS.md, so language, explicit-entry routing, and manuscript-safety invariants track the installed toolkit. Detailed strategy routing lives inask-author; operational procedures remain in the active skills and their references. The CLI checks both agent targets before writing, leaves correct entries unchanged, and atomically replaces other regular files or symlinks. If either agent target is a directory or special file, or any operation fails, stop and report the error rather than recursively deleting anything. This replacement is limited to these two project-root agent files and applies to new projects and existing v3.1 projects. The same invocation creates<project-root>/START_HERE.mdfromassets/START_HERE.mdonly when that path is absent, preserving every existing file, symlink, or directory with that name; this short navigation file points to the manifest, optionalPROJECT_STATE.mdcache, and status check. Complete both actions when the selected branch permits, including after an accepted v3.1 schema with later project-state validation failures. After successful entry-file installation, read<project-root>/AGENTS.mdonce; it is the entry contract for the next strategy. Do not rely on the host to inject it automatically. Then runnode <setup-skill-root>/scripts/setup-report.mjs capture --report-store <report-store> --kind check -- --project-root <project-root> --diagnoseonce, adding the selected branch's strict-execution options to this same call when applicable. This captured read-only check is the final installation and project-state verification: it checks the payload once, verifies both repaired agent entry files, and includes status, the execution-recovery audit, and the cache's format and manifest-freshness checks. Use its command results rather than another directory listing or metadata check of the entry files or registered directories. The report preserves its three independent states: Installation covers payload and project entry files; Project state covers project validation; Strategy gate covers the requested execution gate.Installation: passedis required before describing the toolkit as installed; these results do not requireSTART_HERE.mdto exist or match the installed template. A project-state or strategy-gate failure does not change that installation result. Report project-state failures as project blockers and failed requested gates as blockers for evidence-dependent operations, without repeating doctor or preflight. Do not reproduce a recorded next action from a cache marked stale. - Run
node <setup-skill-root>/scripts/setup-report.mjs render --report-store <report-store>once. It consumes the captured doctor, initialize-or-scaffold, entry-file, and final-check JSON without another project inspection, displays the four labeled sections Installation payload, Project entry files, Manuscript scope, and Execution evidence, and removes the temporary store. Use this rendered report directly rather than reconstructing its fields from the captured JSON. It includes recorded scope hashes and validation, approval and blocker summaries, bounded failure details, and every non-reusable navigation-cache result; a reusable cache adds no detail and no recorded route. Mechanical execution evidence remains distinct from operation completion and current-draft revision convergence. If rendering fails, clean up the report store, report the error, and stop. Apply the remaining semantic routing rules:- Read
../ask-author/SKILL.mdonce and use only the Need labels from its Author-facing commands table as the descriptions for the five strategy entries. - Show the exact host commands, or exact registered skill names on a host with only a skill picker, for
write-fiction,revise-fiction,revise-draft,revise-increment, andrevise-request; forrevise-requestandrevise-increment, apply the selected branch's routing rules for their required scopes and current-draft revision convergence. - Also show the exact host command, or exact registered skill name on a host with only a skill picker, for
ask-authoras the optional "help me choose" routing entry for authors who have not chosen a strategy. - Assume these entries are installed; do not check their host visibility or ask the author to confirm it.
- Explain that the skill the author explicitly starts will confirm the entry operation with the author.
- Wait for the author's choice of a strategy or the
ask-authorrouting entry.
- Read
Boundaries
Create no manuscript prose and fill in no Story Bible creative content. Preserve existing file contents and historical artifacts, except for the two project-root agent entry files described above. START_HERE.md is create-only and never replaces existing content. All five strategies use the same manifest, tracker, reports, sessions, backups, handoffs, registered creation path, and shared references installed with this setup skill. Create no strategy-specific state directory. Setup does not alter Git configuration or .gitignore.
The six independent skills do not require setup for standalone use. Their standalone contracts live in their own skill directories; the installed shared reference payload does not replace those skill-local files.
Use the exact unqualified registered skill names with the current host's skill command or skill picker when displaying the next entries. When the author invoked setup through a host skill command, reuse that command syntax for these entries; on a host with only a skill picker, display the exact registered names. Do not add a qualifier to these unqualified registered skill names.
Report setup as ready only when installation and project state pass, and all selected paths are registered. Assume the five strategy entries are installed without checking or reporting their host visibility. A failed strategy gate blocks its dependent operations without making a passing installation incomplete. Do not start write-fiction, revise-fiction, revise-draft, revise-increment, revise-request, or another writing skill from setup.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/shengchen1998/author-toolkit/setup-author-toolkit">View setup-author-toolkit on skillZs</a>