revise-request
Apply an author-requested revision to existing manuscript from a full-manuscript base scope that reached current-draft revision convergence, reviewing only the request and affected text unless the author requests full-scope review.
How do I install this agent skill?
npx skills add https://github.com/shengchen1998/author-toolkit --skill revise-requestIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill manages a structured workflow for manuscript revisions using local authoring tools. While it processes untrusted text content, it includes mandatory human-in-the-loop approvals and state verification to mitigate risks.
- Socketpass
No alerts
- Snykfail
Risk: CRITICAL · 1 issue
What does this agent skill do?
Revise Request
Coordinate an author-defined revision of existing manuscript content from a full-manuscript base scope that has reached current-draft revision convergence. The request selects the craft work and the default review covers only the declared request-review units. A separate author request for full-scope review is required before claiming that the full-manuscript target scope has reached current-draft revision convergence again. One manifest continues to represent one book. Toolkit skills from one installation are siblings. Resolve <setup-skill-root> as ../setup-author-toolkit relative to this skill's installed directory and use the controlled workflow terminology in <setup-skill-root>/assets/references/terminology.md for shared terms and exact contract values.
Start or resume
- Locate
.author-toolkit/project.json. If it is absent, display the host command forsetup-author-toolkitand stop. Use the same setup command forms asrevise-fiction. If the manifest has any schema version other than3.1.0, report that it is unsupported and stop without changing it. - Read each needed callee's installed
SKILL.mdand applicable references. Execute its procedure directly. A user-invoked entry starts only through its author-facing command, never as an internal callee. - Run the
author-projectprocedure once withcheck --digest, passing--require-sibling author-project --require-sibling revise-fiction --require-sibling revise-draft. Its single JSON result supplies the bounded projection, project validation, ordinary status, and refreshed navigation cache, with installation and entry-file checks included; use this result instead of separate doctor,read,validate,status, or standalone digest-generation calls for this startup. If it reports an incomplete toolkit installation, stop and report the missing installed files; do not deploy contracts into the project. If another check fails, rerun it once with--diagnoseinstead of--digest, report the failure, and stop at the applicable author gate. ReadPROJECT_STATE.mdfirst, then read only the tracker sections, sessions, current plan, scope, pending action, coverage units and index, review jobs, and prior revision evidence needed for the requested work. Do not read the full tracker or backup index for routine navigation. For routine navigation, usePROJECT_STATE.mdpointers, targeted artifact reads, or aggregate counts rather than emitting complete file listings of the registered session or backup directories. Reconcile an interrupted write or commit before dependent work. - On a new request, inventory the complete manuscript and verify that the confirmed base scope is a full-manuscript scope that has reached Current-draft revision convergence. Run
author-project statuswith the base scope ID and--require-execution; a recorded statement of that state does not replace this check. Open findings, missing or stale required coverage, unverified approvals, and scope or hash mismatches prevent entry. When strict execution status fails, record the affected operation asblocked. Showrevise-incrementwhen creation or import grew the scope; otherwise showrevise-draft. Stop rather than using the new request to bypass unfinished work for the base scope. - Record the author's requested outcome, constraints, named paths or passages, and requested checks in the intake report and session note. Establish the full-manuscript target scope, request set, initial work set, and reusable evidence. The target scope always contains the complete current manuscript. The request set contains the text the author asked to change and the checks they requested; the work set also contains dependencies and paths that need new current evidence for the target scope.
- On the first run, select
intake. Present the verified base scope, request interpretation, target scope, request set, work set, operation route, and evidence gaps. Proposereview_mode=requestwith declared changed passages and dependency-affected passages, including necessary local reading context. Ask whether the author confirms that request review or explicitly requests a full-target current-scope review instead. Stop for author confirmation of this request plan. If the confirmed current scope already matches the requested target scope, reuse it as the target scope; otherwise, when membership, order, or exclusions change, useauthor-projectwithrecord-scopeto record the proposed target before presenting it, then confirm that exact unchanged record after explicit author confirmation. Request-plan confirmation does not approve manuscript edits. - On resume, locate the active request in the latest session and any current plan. Treat its verified base scope as historical evidence: the entry's own approved edits are expected to stale that base scope, so rederive the current work set and current-draft revision convergence instead of requiring the edited manuscript to match the base scope. Keep one active request per book and preserve completed request records in sessions and reports.
- Select the earliest operation required by the request or by missing, partial, or stale evidence required for current-draft revision convergence in the sequence below. Read the corresponding gate in
../revise-draft/references/stage-gates.md, plus the rules below. Record the operation throughauthor-projectwithset-pending, run one operation, and stop at its author gate withawaiting_authororblocked.
The current-draft operation sequence is intake, draft-baseline, diagnostics, characters, revision-plan, chapter-revision, prose, ai-cleanup, and current-scope-review. Enter at the operation applicable to the request and reuse current evidence for the base scope rather than automatically rerunning the full sequence. Clean-audit continuation within prose follows prose-mechanics and does not advance to another operation.
Request routing
Translate the requested outcome into ordered craft operations without narrowing the coverage required for current-draft revision convergence. Developmental and continuity changes precede prose and AI cleanup. A combined request may use several routes; an unspecified revision type is not permission to make unrelated edits.
| Operation or requested work | Request-driven work |
|---|---|
intake | Verify that the full-manuscript base scope has reached current-draft revision convergence; record the request, full-manuscript target scope, request set, work set, and reusable evidence |
draft-baseline | Refresh input versions and affected observations, interpretations, and unknowns for the full target scope after approved changes |
diagnostics | Use fiction-workshop for requested or affected scene, causality, pacing, information-release, POV, and chapter-transition work |
| Requested macro structure | Use story-structure Audit as additional structure evidence over the existing text; it does not replace current-draft diagnostics or impose revise-fiction final-review gates |
characters | Recheck requested or affected motivation, voice, arcs, relationships, knowledge states, timeline, continuity, and world facts; use character-archetypes when requested or profiles exist |
revision-plan | Turn the confirmed request and accepted findings into exact batches with paths, actions, input versions, constraints, and acceptance evidence |
chapter-revision | Apply only approved request and review-finding batches with current hashes and backups |
prose | Use prose-mechanics or fiction-workshop line editing for the requested prose outcome, then restore every affected applicable audit |
ai-cleanup | Follow <setup-skill-root>/assets/references/avoid-ai-writing-adapter.md for requested detection or approved rewrites and affected rechecks |
current-scope-review | By default run author-review with review_mode=request over only the declared request-review units and all three perspectives. Only when the author explicitly requests full-scope review, run review_mode=current-scope over the full target scope |
For each diagnostic callee, supply the operation identifier, destination report path, named inputs and versions, target scope_id, BCP 47 language, tracker path, and resolved <setup-skill-root>. Require <setup-skill-root>/assets/references/workflow-reporting.md and finding v2, and supply ../revise-fiction/references/state-and-findings.md resolved from the callee's installed skill directory as the shared state reference. Identify the request-set and work-set paths in the report notes and record actual operation coverage and subchecks. Additional request-specific evidence does not replace any current-draft gate.
New chapters or imported external text belong to write-fiction and then revise-increment. A request may split, merge, delete, or reorder existing manuscript files only through an approved batch and a newly confirmed full-manuscript target scope.
Impact and current-draft revision convergence loop
After every approved change, apply the shared stale-propagation rules to actual text and dependencies. Expand the work set with every path whose required evidence became stale, every path affected by a baseline or plan change, and every target-scope path lacking a reusable full local coverage unit. Preserve unaffected full local coverage units only when operation, source-report identity, paths and hashes, requested subchecks, and dependencies match. Compose a new coverage index for the target scope and a full-target synthesis for every affected operation; never retain an old full-target synthesis because local units survived.
Required coverage for current-draft revision convergence includes every applicable subcheck required by the base scope's current-draft revision convergence evidence plus newly requested checks. A narrower revision request does not silently reduce that required coverage. Partial, skipped, or degraded requested applicable coverage remains a blocker for current-draft revision convergence under the shared definition.
After any manuscript change, run the author-selected review after the latest affected operations. By default, use review_mode=request: select the exact changed passages, the dependency-affected passages, and necessary local reading context as the review units. The scope_id remains the full-manuscript target scope, but review jobs read only those declared units. Do not silently turn them into full chapters or rerun every target-scope path. Complete the request-review units once per logical round after the round's approved edits, rather than immediately after each individual finding. A newly discovered dependency requires author confirmation before adding its passages to the request-review units. Base-scope review reports do not replace changed request-review units.
Only an explicit author request for full-scope review selects a new full-target current-scope review. In that mode, all three perspectives cover the full target scope and the fresh reader reads it sequentially through one continuous reviewer state even when the request names one passage. Review-input changes stale dependent review evidence under the shared rules.
Process review findings through the author decision loop and the shared review-brief decision-reuse rules. An accepted edit returns to revision-plan and chapter-revision, expands the work set and proposed request-review units by actual impact, and requires new affected coverage and the selected review after the round's approved edits. Deferred or dismissed findings retain their explicit decisions. If a request ends with no manuscript or dependency change, record the no-edit decision and reuse current matching review units; new findings still require decisions.
Use the shared Current-draft revision convergence definition at every completion check. Request completion alone does not establish current-draft revision convergence. The requested work reaches request completion when its approved batches and decisions are finished and its selected request review is complete with every finding reconciled through the review brief. review_mode=request never satisfies the full-target current-scope review requirement for current-draft revision convergence, even when every selected unit reports full coverage. Do not create full-target current-scope review coverage units or a full-target current-scope review synthesis from request-review units. Report request completion and the full-manuscript target scope's current-draft revision convergence separately; after request review without a current full-target current-scope review, state that current-draft revision convergence has not been reverified. If the author explicitly requested full-scope review and all other shared conditions are current, apply the shared current-draft revision convergence definition.
Author interaction and manuscript safety
At every author gate, follow the Author decision loop. Before manuscript changes, read <setup-skill-root>/assets/references/context-fit.md and complete its check before presenting an edit proposal. Show the exact operation, request, full-manuscript target scope, work-set paths, actions, input versions, exact candidates, and checked local context. Record explicit approval in the shared tracker, then run author-project with record-backup for every affected file and approval ID. Run complete-write with the exact approved content to publish it, preserve evidence, and record the execution receipt in one invocation; an interrupted batch has no execution receipt until reconciliation and remains unverified. Restore requires separate explicit approval and a safety backup.
Follow Task-based reading before writing a report, changing finding or approval status, propagating stale evidence, or recording a stopping point. Validate newly written reports and artifacts with the installed validator; preserve historical evidence by default or retire only unreferenced audit artifacts under the explicitly enabled bounded policy.
Strategy switching and shared state
write-fiction, revise-fiction, revise-draft, revise-increment, and revise-request use the same manifest, tracker, reports, sessions, backups, handoffs, and creation state. At any stable operation gate, show the exact command for the author's selected strategy and wait. Switching does not reset evidence, resume writes, or recreate shared project state. This entry adds no report v2 stage value, pending-action operation name, manifest field, or separate state tree.
Finish
Finish each invocation with the request and its decisions, base-scope and target-scope IDs, request-set and work-set paths, selected review mode and request-review units, reused full local coverage units, new reports and syntheses, actual coverage, approvals and backups, unresolved findings, request-completion status, current-draft revision convergence status for the full-manuscript target scope and blockers, and the exact next action. Show the write-fiction, revise-fiction, revise-draft, revise-increment, or revise-request command that matches the author's next goal. Use author-handoff for session transfer after shared state is current.
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/revise-request">View revise-request on skillZs</a>