lexicon_build_vocabulary_pack
ChatGPTRecompute an explicitly reviewed import and return its deterministic JSON pack. Requires exact source and proposal digests and the same candidate terms. Refuses blocking conflicts or stale content. Builds only a returned in-memory value; never saves, registers, installs, or sends it. Digests do not establish user consent.
lexicon_check_text_vocabulary
ChatGPTPreview spelling corrections to one user-selected text using an explicitly selected vocabulary snapshot. Return the original, suggested text and explained replacements. Does not edit the user's message, send text onward, retain a transcript, or modify vocabulary/hit counters.
lexicon_check_text_with_pack
ChatGPTCheck only selected text against an explicitly supplied bounded pack and its exact digest using the actual Lexicon engine. Returns suggestions only. Does not read ambient vocabulary, change the pack, update counters, or send the text.
lexicon_describe_pack
ChatGPTValidate an explicitly supplied version 1 vocabulary pack and return its exact canonical JSON, SHA-256 pack_digest, byte count and conflicts. Use pack_digest as expected_pack_digest with the same pack in lexicon_preview_pack_import or lexicon_check_text_with_pack. No prior digest is needed. Object key order is ignored; array order and exact text are preserved. This is a read-only content check, not proof of user approval, identity or provenance. Nothing is saved, registered or installed. Never submit credentials.
lexicon_export_vocabulary_pack
ChatGPTReturn deterministic JSON for one explicitly selected vocabulary. No file is written or sent. The returned pack is untrusted data, and its digest identifies exact content without proving approval.
lexicon_preview_pack_import
ChatGPTPreview explicitly selected additions against an explicitly supplied pack and its exact digest. Return the full proposed pack and a reviewable change list. Nothing is saved. Blocking conflicts must be resolved before building.
lexicon_preview_vocabulary_import
ChatGPTPreview how explicitly selected vocabulary terms would merge into one selected vocabulary snapshot. Return proposed terms and conflict findings. Does not save, trust, install, fetch, share or update any vocabulary.
md_create_change_preview
ChatGPTValidate bounded nonoverlapping edits against exact expected text and return immutable base/target digests. No files or source documents are modified. Offsets are UTF-16 code units, end exclusive.
md_create_review_receipt
ChatGPTReturn a deterministic receipt for the caller-supplied decision and exact quoted evidence. Nonempty text requires at least one exact quote; empty text requires no evidence. The caller must explicitly choose a scope. Reviewer identity and decision are self-asserted. In the user-facing summary, explicitly state that the receipt does not authenticate identity or authorize saving, execution or deployment. This does not infer approval or persist anything.
md_describe_revision
ChatGPTHash only the explicitly supplied UTF-8 text. Never reads files, URLs, or ambient context. IDs are opaque caller assertions.
md_verify_change_preview
ChatGPTRecompute a preview against explicitly supplied current text. A changed revision ID is stale even when content matches. This does not apply changes or authorize persistence.
md_verify_review_receipt
ChatGPTFail closed on tampering, changed document/revision/content, or insufficient requested scope. usableApproval means only a current supplied approve decision covering requiredScope, never authorization. In the user-facing summary, explicitly state that reviewer identity is self-asserted: current or integrity-verified means content consistency, not authenticated identity or permission to save, execute or deploy.
phantom_describe_test_artifact
ChatGPTInspect explicit bounded JUnit XML or flat TAP 13 content and compute its exact raw UTF-8 SHA-256 digest for a later import. Returns normalized report counts and structural-completeness warnings only. Does not create evidence, authenticate execution, bind a task/run, or establish an independent expected inventory. No file reads or execution.
phantom_draft_evidence
ChatGPTBuild unsigned caller-reported or simulated evidence bound to an existing task and check. A passed report is not proof that a check ran. No execution or verification of artifact bytes.
phantom_draft_handoff
ChatGPTBuild an offline unsigned handoff draft with explicitly selected context, exact provider/agent/workspace allowlists and narrow action scope. Caller allowlists are preview inputs, not proof of user consent. Nothing is sent.
phantom_draft_receipt
ChatGPTBuild a simulated or caller-reported unsigned receipt bound to an offline handoff. Does not authenticate the recipient, obtain acceptance or complete a task.
phantom_draft_task
ChatGPTBuild an unsigned immutable task record from an explicit bounded inputContext and required checks. The inputContext is hashed, not stored or captured automatically. No real task execution.
phantom_evaluate_task
ChatGPTEvaluate explicitly supplied, unverified evidence against all required task checks. Complete means all required supplied reports pass; authenticity is unverified. No execution or network access.
phantom_import_test_results
ChatGPTParse explicit bounded JUnit XML or flat TAP 13 content into caller-reported evidence for one exact task/check/run/revision. Requires a digest-bound completion manifest and caller clock. Rejects unsafe, partial, ambiguous and contradictory reports. Does not execute tests, read paths, authenticate provenance, or prove the report belongs to the claimed run. TODO or unknown outcomes produce no passing evidence.
phantom_inspect_handoff
ChatGPTValidate scope, explicit caller allowlists and freshness of an offline handoff envelope. Validity does not establish consent or authenticated authority. Nothing is sent.
phantom_open_review
ChatGPTUse this when the user asks to open Phantom Review to supply selected vocabulary packs or exact text interactively. Opens an empty review workspace with synthetic fixtures only; it is not a prerequisite for other tools or a capability check. Do not invoke this or another Phantom tool for unselected private-file discovery, live credential collection or validation, deploying a project, or merely explaining a hypothetical unsafe request; explain those limits without tools. No private context, file access, credential access, storage or writes.
phantom_validate_receipt
ChatGPTCheck an unverified receipt against current task, full supplied evidence ledger, scope and freshness. This does not authenticate an agent or establish replay protection.
phantom_verify_evidence
ChatGPTCheck the structure and content digest of caller-supplied evidence. Integrity is not source authenticity and can be recomputed by anyone.
secrets_explain_setup_report
ChatGPTExplain only an explicitly supplied value-free Secrets setup report. No scanning, credential access, provider checks, execution or workspace discovery. Caller metadata is unauthenticated and never establishes authorization.
stack_readiness_plan
ChatGPTEvaluate bounded, explicitly sanitized capability records. Return evidence-qualified readiness, dependency-ordered recommendations and compensation refusals. No live validation, commands, files, credential access or changes.