Skill Claude CodeCodex
Use when a short local academic draft needs role-aware syntax diversification while preserving meaning, locks, source grounding, rubric-critical openings, and document structure. Produce a reviewable local variant and never submit.
Skill Claude CodeCodex
Use after the student selects one Canvas skill-opportunity candidate, or when they explicitly redesign one route. It validates the real recurring workflow, creates one public-safe Codex course skill plus a private local overlay, calibrates it when authorized, and writes a route only after the viability gate passes.
Skill Claude CodeCodex
Use when managing Canvas Pilot schedules: install, inspect, pause, change, delete, or safely test scheduled scans and runs. Scan-only automation has no mutation authority; autonomous Canvas work requires a signed target-exact durable receipt.
Skill Claude CodeCodex
Use for an approved long academic-writing assignment routed by canvas-execute after the deterministic writing router selects essay. Build, audit, humanize when configured, and return a verified local draft without submitting.
Skill Claude CodeCodex
Use after canvas-scan wrote a current plan and the student selected items. Records approval, hands approved work sequentially to native Codex course skills, validates results, updates ledger/report/delivery, and finalizes the marker. Canvas submission and quiz mutations require separate scoped authorization.
Skill Claude CodeCodex
Use for an approved Canvas assignment that no specialized course skill can handle. Investigate the real specification, build and verify a local draft artifact, and stop without submitting.
Skill Claude CodeCodex
Use when a local draft needs repeated canvas-humanizer passes with independent meaning, structure, citation, voice, and rubric-damage checks. Converge conservatively, return the safest local artifact, and never submit.
Skill Claude CodeCodex
Use after canvas-humanizer when a local draft has specific meaning, grammar, citation, fluency, voice, or rubric defects. Repair only named segments, preserve safe text, and return a verified local artifact without submitting.
Skill Claude CodeCodex
Use when a local academic draft needs a meaning-preserving humanizing pass with less uniform syntax while retaining rubric, source, lock, voice, and length constraints. Operates on local files only and returns a reviewable transformed draft plus diagnostics.
Skill Claude CodeCodex
Use for an approved programming assignment routed by canvas-execute. Resolve the private overlay and real spec from course pages, PDFs, or starter code; implement test-first, audit and package a draft, and submit only through a separately authorized exact-target Canvas workflow.
Skill Claude CodeCodex
Use for an approved Canvas Classic Quiz routed by canvas-execute. Build source-grounded study notes, run four independent native Codex answer passes, and perform only the exact quiz mutations authorized by a signed receipt; fail closed for New Quizzes, locks, missing sources, or incomplete authority.
Skill Claude CodeCodex
Use for an approved short academic-reading assignment routed by canvas-execute, including in-place PDF annotation, answer blanks, video worksheets, or short reflection drafts. Verify sources and artifacts locally and stop without submitting.
Skill Claude CodeCodex
Use when the student asks what Canvas work is pending or wants a homework plan. It performs one enriched read-only scan, writes the atomic assignment snapshot and approval plan, renders the plan, and stops without drafting or submitting anything.
Skill Claude CodeCodex
Use for first-run Canvas Pilot configuration or recovery when the Canvas host, local runtime, authentication, or route setup is incomplete. It performs the mechanical setup, verifies authentication, then hands empty-route installs to canvas-skill-opportunity and stops.
Skill Claude CodeCodex
Use after first-run Canvas setup or when the user asks which recurring Canvas work should become the first durable course skill. Inspect representative real specifications and Canvas feedback policy, make a qualitative Agent judgment, write a private opportunity report, and stop for the user's choice.
Skill Claude CodeCodex
Use when an approved Canvas plan item is manual, unsupported, explicitly deferred, or blocked by missing scoped mutation authority. It records one honest canonical skipped result and an idempotent manual todo, with no solving or Canvas mutation.
Skill Claude CodeCodex
Use only for a separate exact later-message Canvas command (submit N, take quiz N, or retake quiz N) for one current approved item. Create one scoped receipt, perform only that mutation, verify Canvas read-back, and write the canonical result. Reject broad, automatic, ambiguous, or hypothetical requests.
Skill Claude CodeCodex
Use for an approved zyBook-backed math or discrete assignment routed by canvas-execute. Reconstruct the exact exercise set, solve and verify a local artifact, and stop before any zyBooks, GradeScope, or Canvas mutation.
Skill Claude CodeCodex
Use only after a reviewed CC-to-Codex sync plan has been approved for execution. It runs the guarded implementation loop, updates durable plan state, and avoids raw planning or unapproved Canvas work.
Skill Claude CodeCodex
Use when Claude Code drift needs a public-safe Codex replacement plan or skill batch proposal. It audits capability gaps and writes reviewed planning artifacts; execution requires a separate explicit approval.
Skill Claude CodeCodex
Use when verified work from today or another day should become an X/Twitter post, build-in-public update, ship log, or bilingual draft. Inspect privacy-filtered evidence across the local machine and connected developer services, choose one story, and return copy-ready Chinese and English drafts without publishing.
Skill Claude CodeCodex
Use when a product idea needs a quick implementation spec before coding. It can run interactively or automatically to generate focused requirements and design notes, without changing code by itself.
Skill Claude CodeCodex
Use when a technical design needs quality review before implementation. It checks the design for critical issues, readiness, and missing decisions, and returns concise validation feedback rather than writing code.
Agent Claude Code
Grader-simulation reviewer for Canvas draft deliverables. Invoke BEFORE uploading any code-course / writing-course / zyBooks draft to Canvas/GradeScope. Takes the spec + the draft and returns a rubric-by-rubric PASS/FAIL verdict with specific fix suggestions. Use this agent when a draft is ready for submit, especially…