Continuation

wep task continuation TASK-001 [--json] reads a task’s optional continuation.json. The record travels with the repository if you commit it. The command does not create, update, import or migrate records, resume work, or run commands. Existing tasks need no new files.

What the result means

status	Meaning	Exit code
missing	The task exists but has no sidecar. Execution is unknown, not “never started”.	0
valid	The record satisfies schema v1 and names the requested task directory.	0
invalid	The sidecar exceeds the size limit or contains invalid UTF-8, JSON or schema data.	1
unsupported	The JSON declares another integer schema version. It is not interpreted.	1

Invalid task IDs, duplicate or missing task directories, blocked paths, symlinks, non-regular files and other filesystem/policy errors fail explicitly. They are not presented as a missing legacy record. The command reads at most 64 KiB plus one byte to detect an oversized sidecar.

JSON results contain taskId, path, status, executionState, permission, verification, continuationDigest, record and reasons. record and continuationDigest are null unless status is valid. The digest hashes the original sidecar bytes; whitespace changes it.

Valid is structural, not verified or authorized. permission and verification always remain not-evaluated. executionState is the producer’s declaration. In particular, a persisted active state does not authorize or restart a process. The reader does not compare the declared capsule digest or code content with the current workspace.

Snapshot availability: "verified" and validation summary status: "complete" are also producer declarations. They do not establish who ran tests or whether the results apply now. The reader never executes a command found in a record.

Minimal v1 example

For .wep/tasks/TASK-001-example/continuation.json:

{
  "schemaVersion": 1,
  "task": {
    "taskId": "TASK-001",
    "capsulePath": ".wep/tasks/TASK-001-example",
    "capsuleDigest": "sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
    "capsuleRevision": { "gitCommit": null }
  },
  "latestExecution": null,
  "history": []
}

The repeated a digest is an illustrative placeholder with a valid shape, not evidence of a real capsule revision. A producer must supply the actual digest. The directory and ID must match the task being queried. No five-file migration or automatic execution initialization occurs.

Execution fields

When latestExecution is not null it contains these required fields:

Field	Value
runId	run_ plus 1–80 ASCII letters, digits, _ or -.
state	planned, ready, active, blocked, handoff-requested, completed, abandoned, superseded.
workspace	gitRemote (a non-sensitive label or unknown), commit (null or Git hash), content.
owner	kind: orchestrator or manual-session; id: label or unknown.
model	requested and observed strings; use unknown for information not provided.
progress, blockers, decisionsPending	Arrays of short summaries.
attempts	Portable budget declaration described below.
validationSummary	Declared command outcomes described below.
resumption	requiresReauthorization must be true; a non-empty reason is required.

workspace.content is either { "kind": "unavailable" } or an object with kind: "judge-snapshot", snapshotId, reviewContentDigest and availability (verified or unavailable). Snapshot IDs start with SNAPSHOT: followed by 1–128 ASCII letters, digits, _ or -. reviewWorkspaceDigest is deliberately not a portable field: filesystem integrity of a particular local snapshot is different from content identity across checkouts.

attempts contains budgetAccountId (budget_ plus 1–80 identifier characters), consumed (a non-negative safe integer or unknown), limit (a positive safe integer), provenance (portable-summary, legacy-local-log, unknown) and localReconciliation (not-available, reconciled, conflict). A count above the limit is retained as exhausted history, not silently clamped. Reading without local logs neither resets the budget nor corroborates it. The reader does not reconcile local logs or reserve an attempt.

validationSummary contains contractDigest, codeReviewContentDigest (digest or null), required (an array of { "command": "…", "status": "passed|failed|not-run" } objects), and status (unknown, incomplete, complete, not-applicable-to-runtime). The notation passed|failed|not-run lists alternatives; choose one in a real record.

A structurally complete summary needs a code content digest and at least one required command, all marked passed. not-applicable-to-runtime requires an empty required-command list and an additional non-empty reason. No other status accepts reason in that object.

history contains at most 20 execution objects, all in terminal states (completed, abandoned, superseded). Run IDs must be unique and differ from the latest execution. Oversized history is rejected, never truncated on read.

Limits, privacy and compatibility

Schema v1 rejects unknown fields at every object level. Free strings are at most 1,024 characters; ordinary arrays are at most 100 entries. Free strings must be non-empty; explanatory reason fields must also contain non-whitespace text. Digests use sha256: plus 64 lowercase hexadecimal characters. Git commits are null or 40/64 lowercase hexadecimal characters. Numbers and enums are not coerced from strings or arrays.

The sidecar should contain only non-sensitive summaries. Do not put transcripts, raw logs, credentials, private URLs, grants or personal comprehension answers in it. Closed schema objects prevent adding arbitrary permission/log fields, but cannot detect secrets embedded in free text. Producers are responsible for reviewing what they make portable. JSON output includes the valid record; terminal output shows only structural status and reasons.

Judge boundary and execution metadata

A regular continuation.json accepted as valid for the reviewed task is excluded from the snapshot’s reviewContentDigest and reviewWorkspaceDigest. Its bytes remain in the full contentDigest and workspaceDigest; modifying a sidecar inside an immutable snapshot still fails integrity verification. An invalid, unsupported or irregular sidecar is not excluded from reviewable content.

wep judge current separates the whole workspace status from reviewBoundary and reports executionMetadata as ABSENT, UNCHANGED, CREATED, REMOVED or ADVANCED. Creating, removing or updating only a valid sidecar can make whole-workspace status NEWER_CHANGES while leaving reviewBoundary CURRENT. A code change alongside it still changes the review boundary. These labels describe content comparison; they do not authorize resumption or prove the sidecar’s execution claims. See Judge.

The five canonical capsule documents and their digest calculation are unchanged.