`temper resource`

Manage resources (tasks, goals, sessions, research, concepts, decisions)

Manage resources (tasks, goals, sessions, research, concepts, decisions)

Usage: temper resource [OPTIONS] <COMMAND>

Commands:
  create              Create a new resource
  list                List resources, optionally of a given type
  describe-open-meta  Describe the recognized open_meta conventions (the self-describing schema)
  doc-types           List the document types, with whether each has a schema and its required fields
  describe-type       Describe one document type: its JSON Schema, required fields, closed vocabularies (a task's stages, a goal's statuses), and an example managed tier
  connections         List a resource's connections: its edges in both directions, one page
  citation-audits     List the signed audit verdicts recorded against a finding's citations
  show                Show a resource's content
  evidence            Show a resource's evidential-standing shape — the maturity vector (independence-discounted breadth, adversarial survival, contradiction balance, freshness) plus a lossy read-time `band` chip carried WITH the shape, never in place of it. Calls GET /evidence
  read-block          Read one content block by address — the three-state resolution
  update              Update a resource's frontmatter and/or body
  annotate            Attach provenance sources to a resource's block — WITHOUT a body revise (issue #355)
  meta                Read or replace a resource's frontmatter without touching the body — the dedicated metadata-only door (`GET/PUT /api/resources/{id}/meta`)
  audit-citation      Record a citation-audit verdict against one (block, source) citation — the block-addressed door (`POST /api/citation-audits`)
  delete              Delete a resource (soft-delete via the API)
  reassign            Reassign a resource's owner (mis-attribution self-fix, or a team admin acting over a resource scoped to their team)
  grant               Grant a capability on a resource to a profile or team (system-admin, a can_grant holder, or the resource owner)
  revoke              Revoke a capability grant on a resource (system-admin, a can_grant holder, or the owner)
  facet               Set a facet property on a resource (cloud-mode-only API write)
  facets              List the live facets of a resource — one row per assert, with weights
  help                Print this message or the help of the given subcommand(s)

Options:
      --vault <VAULT>      Path to vault (overrides TEMPER_VAULT and auto-detection)
      --format <FORMAT>    Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default
      --embed-threads <N>  ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1
      --color <COLOR>      Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto
  -h, --help               Print help

`temper resource create`

Create a new resource

Usage: temper resource create [OPTIONS] --type <TYPE>

Options:
      --type <TYPE>
          Resource type (task, goal, session, research, concept, decision)

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --title <TITLE>
          Resource title

      --context <CONTEXT>
          Context ref (UUID or @owner/slug, e.g. @me/temper or +team/general). Mutually exclusive with --cogmap; specify exactly one home

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --cogmap <COGMAP>
          Cognitive-map ref (UUID or decorated `slug-<uuid>`) to home the resource in. Mutually exclusive with --context; specify exactly one

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --mode <MODE>
          Work mode: plan or build (task only)

      --effort <EFFORT>
          Work effort: small, medium, large (task only)

      --open-meta <OPEN_META>
          Open (caller-defined) frontmatter as a JSON object string, e.g. --open-meta '{"marker":"x","reviewed":true}'. These are the free-form "bring-your-own" fields; the closed temper-* vocabulary uses the typed flags (--mode/--effort/…). Must be a JSON object; a null value is refused on create (null deletes a key on update — omit the key here)

      --goal <GOAL>
          Link this resource to a goal by ref (UUID or decorated `slug-<uuid>`). Projects a live `advances`→goal edge from the new resource on create

      --task <TASK>
          Link this session to a task by slug (session only). Asserts a session→task `advances` relationship after creation

      --show-template
          Print the raw template and exit

      --body <BODY>
          Body content: '@PATH' reads a file, '-' reads stdin, or omit to use piped stdin implicitly

      --from <FROM>
          Source path or http(s) URL — extract markdown via temper-ingest and use as body. Supported formats: md/markdown, txt/text, html/htm, pdf (text-layer PDFs only — a scanned or image-only PDF extracts nothing, as there is no OCR). Mutually exclusive with --body. http(s) URLs are fetched; a local file may be given as a plain path or a file:// URI

      --sources <SOURCES>
          Provenance sources this body was distilled from — comma-separated resource refs (UUID or decorated) and/or external http/https URLs. Each becomes a block-provenance record on the resource's body block (URLs via the 'remote' kind)

      --sources-as-edges
          Also assert a `derived_from` edge from the new resource to each resource-valued `--sources` entry. Remote URLs are skipped (no edge target).
          
          Not atomic: the edges are asserted after the create commits. A failed edge warns rather than failing the command — `edge assert` is idempotent, so re-asserting is safe, while re-running a create is not.

      --no-source
          Suppress the `--from <url>` provenance default. By default a URL `--from` sets the resource's origin and seeds a Remote block-provenance record from it (so `create --from <url>` is citation-grade with no extra flags); `--no-source` opts out, leaving the origin empty and recording no provenance. Mutually exclusive with `--sources`

      --preserve-source
          Preserve the original as a blob (`derivation-source-preservable`, D3): with a LOCAL file `--from`, the file's raw bytes are committed as a blob homed in the same context/cogmap, and a `derivation_source` edge is asserted from the resource to the blob — the provenance question later is a graph read (`blob relations`/`blob get`). Not atomic: blob + edge assert after the create commits, warnings rather than failures (re-running is safe on the blob; the create is not re-runnable)

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource list`

List resources, optionally of a given type

Usage: temper resource list [OPTIONS]

Options:
      --type <TYPE>
          Resource type (task, goal, session, research, concept, decision). Optional — omit to list across every doc type, which is what makes a cross-type axis like `--tag` answerable in one call
      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)
      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default
      --tag <TAG>
          Filter by tag (repeatable). `--tag a --tag b` returns resources carrying BOTH — each added tag narrows. Matching is exact and case-insensitive. Not doc-type-scoped: tags span every doc type, so this composes with `--type` or stands alone
      --context <CONTEXT>
          Filter by context ref (UUID or @owner/slug, e.g. @me/temper or +team/general)
      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1
      --cogmap <COGMAP>
          Scope to resources homed in one or more cognitive maps (UUID or decorated ref). Repeatable — `--cogmap A --cogmap B` lists resources homed in either. Mutually exclusive with --context
      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto
      --ids <IDS>
          List these resources by id (UUID or decorated ref; comma-separated or repeated), at most 256. Returns the ones you can see and nothing else: an id that does not exist, is deleted, is not visible to you, or is an unfinished upload is simply absent from the rows. Raises the page size to at least the number of ids
      --limit <LIMIT>
          Page size (default 20). A DEFAULT, not a cap: whatever you pass is honoured unchanged, and there is no server-side clamp. The response always carries `total` (the full match count), `returned`, and `truncated`, so a capped page is self-evident. Conflicts with --all
      --all
          Return ALL matching results (no page cap). Reach for this before asserting a set is complete or a resource is absent. Conflicts with --limit
      --offset <OFFSET>
          Skip the first N matching results (pagination). Conflicts with --page, which is the same axis counted in pages instead of rows
      --page <PAGE>
          Page number, 1-indexed — `--page 1` is the first page. Resolves to `(page - 1) * <effective limit>`, so it counts in whatever `--limit` is in force (`--page 3 --limit 5` starts at row 10, not 40). Conflicts with --offset (the same axis in rows) and with --all (an uncapped page has no page number)
      --sort <SORT>
          Sort as `<field>[:asc|desc]`. Fields: updated, created, title, stage, seq, context, doctype. Direction defaults per field (time/seq → desc, text → asc). Omit for the default `updated:desc`
      --title-contains <TITLE_CONTAINS>
          Filter to titles containing this substring (case-insensitive). A cheap way to narrow a large set instead of paging blind
      --stage <STAGE>
          Filter by stage (task only)
      --goal <GOAL>
          Filter to resources advancing this goal (any type: its tasks, decisions, evidence)
      --status <STATUS>
          Filter by status (goal only)
      --with <WITH>
          Add a section to every row (comma-separated or repeated). `--with open-meta` fills the open metadata tier — the same envelope and the same row type as the default list, since asking for a section adds a part to the one shape rather than selecting a second one. The managed tier is always present either way. `body` is deliberately not offered here: a page of reconstructed bodies is an unbounded payload behind a flag that reads as cheap — use `show` per row [possible values: open-meta, embedding-status]
      --without <WITHOUT>
          Drop a section from every row (comma-separated or repeated). `list` asks for none by default, so this is only meaningful against a `--with` on the same invocation — and naming one section in both is a hard error, not a precedence rule [possible values: open-meta, embedding-status]
      --fields <FIELDS>
          Subselect top-level response keys on each row (anchor key always preserved). Use jq for nested projection
  -h, --help
          Print help

`temper resource describe-open-meta`

Describe the recognized open_meta conventions (the self-describing schema)

Prints the recognized open (caller-defined) frontmatter keys, their shapes, and — via each key's description — whether it is FTS-indexed (and at what weight) or shape-only, plus the discouraged bare keys. The open tier stays free-form; this is guidance, not a closed vocabulary. Mirrors the MCP `describe_open_meta` tool.

Asks the server (`GET /api/schema/open-meta`): the deployment you write to is the one that validates, and its schema can differ from this binary's when the two versions do. `--local` answers offline from the schema compiled into this binary.

Usage: temper resource describe-open-meta [OPTIONS]

Options:
      --local
          Answer from the schema compiled into this binary instead of asking the server

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource doc-types`

List the document types, with whether each has a schema and its required fields.

Asks the server (`GET /api/schema/doc-types`); `--local` answers offline from this binary.

Usage: temper resource doc-types [OPTIONS]

Options:
      --local
          Answer from the schema compiled into this binary instead of asking the server

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource describe-type`

Describe one document type: its JSON Schema, required fields, closed vocabularies (a task's stages, a goal's statuses), and an example managed tier.

Asks the server (`GET /api/schema/doc-types/{name}`); `--local` answers offline from this binary.

Usage: temper resource describe-type [OPTIONS] <NAME>

Arguments:
  <NAME>
          The document type, e.g. `task`

Options:
      --local
          Answer from the schema compiled into this binary instead of asking the server

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource connections`

List a resource's connections: its edges in both directions, one page.

Wraps `GET /api/resources/{id}/connections`; the answer says whether rows were left out.

Usage: temper resource connections [OPTIONS] <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --limit <LIMIT>
          Max connections (the service defaults to 50 and clamps to 1..=200)

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource citation-audits`

List the signed audit verdicts recorded against a finding's citations.

Wraps `GET /api/resources/{id}/citation-audits`. Record one with `audit-citation`.

Usage: temper resource citation-audits [OPTIONS] <REF>

Arguments:
  <REF>
          The finding's ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource show`

Show a resource's content

Output formats. `--format json` returns one document whose `content` key is `GET /api/resources/{id}/content`'s markdown unchanged (`--fields content` keeps the envelope and `id`), so `content` is byte-exact when `body_storage` is `verbatim`. `toon` (the default on a terminal) prints the frontmatter, then `---\n`, then the body plus one extra trailing `\n`: it is not byte-exact, so never hash or diff it.

Usage: temper resource show [OPTIONS] <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --edges
          Show graph edges connected to this resource

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --lineage
          Show the resource's derived_from lineage — what it derives from (ancestors) and what derives from it (descendants), access-gated. Calls GET /lineage

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --provenance
          Show itemized per-block provenance — the sources each of the resource's content blocks was distilled from. Calls GET /provenance

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --with <WITH>
          Add a section (comma-separated or repeated). `show` already carries `body` and `open-meta`; `--with edges` folds this resource's graph edges into the same document (the long form of `--edges`).
          
          The peer is polymorphic (the 2026-09-02 S6 deferral, landed): resource peers carry their title; blob peers — the `derivation_source` edge `--preserve-source` asserts, say — are addressed by bare id alone, so this section answers "what is this resource derived from" from the resource side. `temper blob relations <blob-id>` remains the blob-side view. Cogmap-ended edges are not rendered here, and the walk surfaces stay node-typed: an edge listing may render a blob, a walk never materializes one.
          
          [possible values: body, open-meta, edges, embedding-status, byte-blocks]

      --without <WITHOUT>
          Drop a section (comma-separated or repeated). `--without body` is the cheap orientation read: everything `show` returns except the reconstructed markdown, and it composes freely with `--with edges`. Naming one section in both `--with` and `--without` is a hard error, not a precedence rule
          
          [possible values: body, open-meta, edges, embedding-status, byte-blocks]

      --fields <FIELDS>
          Subselect top-level response keys (the anchor key `id` is always preserved). Use jq for nested projection

  -h, --help
          Print help (see a summary with '-h')

`temper resource evidence`

Show a resource's evidential-standing shape — the maturity vector (independence-discounted breadth, adversarial survival, contradiction balance, freshness) plus a lossy read-time `band` chip carried WITH the shape, never in place of it. Calls GET /evidence

Usage: temper resource evidence [OPTIONS] <REF>

Arguments:
  <REF>  Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --vault <VAULT>      Path to vault (overrides TEMPER_VAULT and auto-detection)
      --format <FORMAT>    Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default
      --embed-threads <N>  ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1
      --color <COLOR>      Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto
  -h, --help               Print help

`temper resource read-block`

Read one content block by address — the three-state resolution

States `live`, `folded`, or `absent` BY NAME in the envelope's `state` field. `live` carries the block's identity, its chunks' identities (structure and hashes, never prose), and its provenance rows. `folded` means a re-partition folded it away; the envelope carries the persisted attribution history plus a `disposition` stating where the content went — `located` names the absorber and carried successor blocks (only those you can read, each with its `home_resource_id`), `content_gone`, or `unrecorded` (the ledger does not carry the mapping). Address a folded successor by its own composed address `<home>#<block-uuid>`. `absent` — a block that does not exist or is not visible to you, indistinguishable by design — exits non-zero. Calls GET /resources/{id}/blocks/{block_id}.

Usage: temper resource read-block [OPTIONS] <REF> [BLOCK_ID]

Arguments:
  <REF>
          Resource ref: a UUID, the decorated `slug-<uuid>` form, or the composed block address `<resource>#<block-uuid>`

  [BLOCK_ID]
          The content block UUID. Omit when the ref carries the composed form `<resource>#<block-uuid>`

Options:
      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource update`

Update a resource's frontmatter and/or body

Mutates frontmatter from flag args. Optionally rewrites the body — the body trio (content + content_hash + chunks_packed) is PATCHed alongside any frontmatter changes in a single API call.

Body-source precedence, first match wins:

1. `--body @<path>` — read the file; stdin is ignored entirely. 2. `--body -` — read stdin explicitly (blocks; errors on a TTY or empty input). 3. implicit stdin — read stdin only when it is a non-TTY with input *ready* and non-empty (the `cat new.md | temper resource update <ref>` case); an idle or empty pipe is "no body". 4. none of the above — the body is left unchanged; only frontmatter is PATCHed.

FOOTGUN: implicit non-TTY stdin is a body rewrite. Do NOT run `update` inside a redirected loop (`while read n ref; do temper resource update "$ref" --title …; done < refs.txt`): every `update` inherits the loop's stdin (`refs.txt`) and rewrites the body with the leftover lines. For frontmatter-only edits (e.g. `--title`), invoke once per resource with stdin untouched; rewrite a body only via an explicit `cat file | temper resource update <ref>`.

Usage: temper resource update [OPTIONS] <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --type-to <TYPE_TO>
          New resource type (converts the resource)

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --context-to <CONTEXT_TO>
          Move resource to a new context

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --title <TITLE>
          Update title

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --tags <TAGS>
          Add tag, keeping existing ones (repeatable). To replace or clear the list instead, use --open-meta '{"tags":[...]}' (or [] to clear)

      --aliases <ALIASES>
          Add alias, keeping existing ones (repeatable)

      --relates-to <RELATES_TO>
          Add relates-to reference, keeping existing ones (repeatable)

      --references <REFERENCES>
          Add reference, keeping existing ones (repeatable)

      --depends-on <DEPENDS_ON>
          Add depends-on reference, keeping existing ones (repeatable)

      --extends <EXTENDS>
          Add extends reference, keeping existing ones (repeatable)

      --preceded-by <PRECEDED_BY>
          Add preceded-by reference, keeping existing ones (repeatable)

      --derived-from <DERIVED_FROM>
          Add derived-from reference, keeping existing ones (repeatable)

      --open-meta <OPEN_META>
          Open (caller-defined) frontmatter as a JSON object string, e.g. --open-meta '{"marker":"x","reviewed":true}'. REPLACES each key it names — including lists, so '{"tags":[]}' clears tags — and an explicit **null value DELETES the key**, e.g. --open-meta '{"word_count":null}' removes it (on create a null is refused: omit the key). The repeatable flags above ADD instead; when both name a key, this replace lands first and the additions union on top. Free-form "bring-your-own" fields; temper-* keys use the typed flags. Must be a JSON object

      --open-meta-add <OPEN_META_ADD>
          Open (caller-defined) frontmatter to ADD, as a JSON object string of list-valued keys, e.g. --open-meta-add '{"reinforced":["2026-08-02"]}'. Mirrors --open-meta exactly but UNIONS each key over the stored list instead of replacing it, so accumulated history survives. This is the only way to add to a key the repeatable flags above do not name; every value must be a list (a scalar is refused rather than replacing the stored list). Where both this and a repeatable flag name one key, the two sets union

      --stage <STAGE>
          Task stage (backlog, in-progress, done, cancelled)

      --mode <MODE>
          Task mode (plan, build)

      --effort <EFFORT>
          Task effort (small, medium, large)

      --seq <SEQ>
          Task sequence number

      --branch <BRANCH>
          Git branch

      --pr <PR>
          Pull request URL

      --goal <GOAL>
          Set (or replace) the resource's goal by ref (UUID or decorated `slug-<uuid>`). Folds any existing `advances`→goal edge and asserts the new one. Conflicts with --clear-goal

      --clear-goal
          Clear the resource's goal — retract its `advances`→goal edge, leaving it goal-less. Conflicts with --goal

      --status <STATUS>
          Goal status (active, completed, paused, cancelled)

      --body <BODY>
          Body source, first match wins: `@<path>` reads a file (stdin ignored); `-` reads stdin explicitly (blocks; errors on a TTY or empty input); omit to auto-detect a *ready* non-TTY stdin pipe (an idle/empty pipe = no body). WARNING: implicit stdin is a body rewrite — never run this inside a `while read … done < file` loop (each call inherits the redirected file as its body); for frontmatter-only edits leave stdin untouched

      --sources <SOURCES>
          Provenance sources this body was distilled from — comma-separated resource refs (UUID or decorated) or http(s) URLs. Each becomes a block-provenance record on the addressed block. Requires a body update

      --content-block <CONTENT_BLOCK>
          Which content block the body revise + `--sources` target (a block UUID). Omit to address the resource's sole body block (the default); required to revise a resource that has more than one block. The block must belong to the resource and be non-folded

      --expect <JSON>
          Write only if these meta values hold now, as a JSON object: '{"open":{"KEY":VALUE},"managed":{"temper-KEY":VALUE}}'. A top-level null means absent, and a key need not be one this write sets. If any does not hold, nothing is written, the exit code is 1, and `--format json` prints code PRECONDITION_FAILED with each failed key and its current value in `details.failed`. Guard on a value unique to you: a current value equal to your own means an earlier attempt of yours landed

      --expect-absent <TIER:KEY>
          Write only if this key is absent now: `open:<key>` or `managed:<key>`. Repeatable, and combined with --expect; the same as naming the key with null there. A misspelled key reads as absent, so its guard always holds

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource annotate`

Attach provenance sources to a resource's block — WITHOUT a body revise (issue #355).

The annotate-only backfill: records block-provenance rows on the addressed block without re-chunking or re-embedding (body_hash and embeddings are unchanged), so a corpus imported without sources can be made citation-grade cheaply. Verify with `resource show --provenance`.

Span locators ride the source URL verbatim via a URL-fragment convention — e.g. `--sources 'https://example.com/doc.md#L120-L180'` records the line range and surfaces it in `--provenance` output (no schema change; the fragment is preserved end-to-end).

Usage: temper resource annotate [OPTIONS] --sources <SOURCES> <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --sources <SOURCES>
          Provenance sources to attach — comma-separated resource refs (UUID or decorated) or http(s) URLs (optionally with a `#L<start>-L<end>` locator fragment). At least one required. Each becomes a block-provenance record on the addressed block

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --content-block <CONTENT_BLOCK>
          Which content block to annotate (a block UUID). Omit to address the resource's sole body block (the default); required for a resource that has more than one block. The block must belong to the resource and be non-folded

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource meta`

Read or replace a resource's frontmatter without touching the body — the dedicated metadata-only door (`GET/PUT /api/resources/{id}/meta`).

Distinct from `resource update`: that PATCHes frontmatter from typed flags and can carry a body revise; this door names both tiers at once and merges per key.

Usage: temper resource meta [OPTIONS] <COMMAND>

Commands:
  get   Read a resource's frontmatter — both tiers, body untouched
  set   Replace a resource's frontmatter — the metadata-only PUT, both tiers named at once
  help  Print this message or the help of the given subcommand(s)

Options:
      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource meta get`

Read a resource's frontmatter — both tiers, body untouched.

404 when the resource is absent or unreadable.

Usage: temper resource meta get [OPTIONS] <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource meta set`

Replace a resource's frontmatter — the metadata-only PUT, both tiers named at once.

Merge semantics (observed at the backend, not inherited from the route's comment): every key you name is overwritten, every key you omit is preserved — a named open-tier key set to null reads back as absent. No body revise, no re-chunk, no re-embed. 403 when the resource is readable but not modifiable; 404 when absent or unreadable.

Usage: temper resource meta set [OPTIONS] --managed <MANAGED> --open <OPEN> <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --managed <MANAGED>
          Managed (temper-) frontmatter as a JSON object string — the closed vocabulary only; an unknown temper-* key is refused client-side, there is no catch-all. Named keys overwrite; omitted keys are preserved. NOTE: a managed key set to null is a silent NO-OP (deserialized to absent → preserved by the backend merge); there is no clear channel on the managed tier

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --open <OPEN>
          Open (caller-defined) frontmatter as a JSON object string. Named keys overwrite; omitted keys are preserved (the additive channel is `resource update --open-meta-add`). A named key set to null reads back as absent

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --expect <JSON>
          Write only if these meta values hold now, as a JSON object: '{"open":{"KEY":VALUE},"managed":{"temper-KEY":VALUE}}'. A top-level null means absent, and a key need not be one this write sets. If any does not hold, nothing is written, the exit code is 1, and `--format json` prints code PRECONDITION_FAILED with each failed key and its current value in `details.failed`. Guard on a value unique to you: a current value equal to your own means an earlier attempt of yours landed

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --expect-absent <TIER:KEY>
          Write only if this key is absent now: `open:<key>` or `managed:<key>`. Repeatable, and combined with --expect; the same as naming the key with null there. A misspelled key reads as absent, so its guard always holds

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource audit-citation`

Record a citation-audit verdict against one (block, source) citation — the block-addressed door (`POST /api/citation-audits`).

The authorization subject is the finding that owns the block, resolved server-side — the caller never names a finding. Append-only: a later audit never erases an earlier one. Only Resource-kind sources are auditable; the value is the signed defensibility verdict in [-1.0, 1.0]. 404 collapses unreadable finding / unreadable cited source / self-audit / absent block into one sentence; 409 when --invocation names a closed run.

Usage: temper resource audit-citation [OPTIONS] --source <SOURCE> --value <VALUE> <BLOCK>

Arguments:
  <BLOCK>
          The audited citation's block id (from `resource show --provenance`)

Options:
      --source <SOURCE>
          The cited source — a resource ref (UUID or decorated `slug-<uuid>`). Only resource-kind sources are auditable

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --value <VALUE>
          Signed verdict in [-1.0, 1.0]: how much this source supports the specific connection the citation claims — never whether the claim is true

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --reason <REASON>
          Optional free-text rationale, recorded on the ledger row

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource delete`

Delete a resource (soft-delete via the API).

Sets `is_active = false` server-side; the row is preserved. Removing a projected file from disk with `rm` is just a local cache miss and has no server effect — run `temper resource delete` to actually delete, then `temper pull <context>` to re-materialize state on a fresh device. Delete is non-interactive on all surfaces — there is no confirmation prompt. `--force` is vestigial (a no-op holdover from the pre-cloud local-mode TTY gate); it is accepted for clarity but changes nothing.

Usage: temper resource delete [OPTIONS] <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --force
          Skip the local-file confirmation prompt

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource reassign`

Reassign a resource's owner (mis-attribution self-fix, or a team admin acting over a resource scoped to their team).

Sends a `POST /api/resources/{id}/reassign` request via `temper-client`.

Usage: temper resource reassign [OPTIONS] --to <TO> <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --to <TO>
          Recipient profile UUID

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

`temper resource grant`

Grant a capability on a resource to a profile or team (system-admin, a can_grant holder, or the resource owner)

Usage: temper resource grant [OPTIONS] <REF>

Arguments:
  <REF>  Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --to-profile <TO_PROFILE>  Grant to this profile (UUID). Mutually exclusive with `--to-team`
      --vault <VAULT>            Path to vault (overrides TEMPER_VAULT and auto-detection)
      --format <FORMAT>          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default
      --to-team <TO_TEAM>        Grant to this team: a team slug (optionally `+`-prefixed), a decorated `slug-<uuid>` ref, or a team UUID. Mutually exclusive with `--to-profile`
      --embed-threads <N>        ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1
      --read                     Grant read
      --color <COLOR>            Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto
      --write                    Grant write (implies read)
      --grant                    Grant delegated-grant authority (implies read)
  -h, --help                     Print help

`temper resource revoke`

Revoke a capability grant on a resource (system-admin, a can_grant holder, or the owner)

Usage: temper resource revoke [OPTIONS] <REF>

Arguments:
  <REF>  Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --from-profile <FROM_PROFILE>  Revoke this profile's grant (UUID). Mutually exclusive with `--from-team`
      --vault <VAULT>                Path to vault (overrides TEMPER_VAULT and auto-detection)
      --format <FORMAT>              Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default
      --from-team <FROM_TEAM>        Revoke this team's grant: a team slug (optionally `+`-prefixed), a decorated `slug-<uuid>` ref, or a team UUID. Mutually exclusive with `--from-profile`
      --embed-threads <N>            ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1
      --color <COLOR>                Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto
  -h, --help                         Print help

`temper resource facet`

Set a facet property on a resource (cloud-mode-only API write).

Sends a `POST /api/facets` request via `temper-client`.

Usage: temper resource facet [OPTIONS] --values <VALUES> <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --values <VALUES>
          The facet's typed value payload, as a JSON string

      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --weight <WEIGHT>
          Facet weight (default: 1.0)

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --invocation <INVOCATION>
          Correlate this act with an open invocation envelope (its ref/UUID from `invocation open`)

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

      --correlation <CORRELATION>
          Stitch this write into an act-grain thread shared with other writes (a bare UUID you mint). Provenance only — it never authorizes. Omit and the event self-roots

      --confidence <CONFIDENCE>
          Graded authorship confidence: tentative, probable, or confident
          
          [possible values: tentative, probable, confident]

      --reasoning <REASONING>
          Free-text reasoning for the act (authorship; requires --confidence)

      --rationale <RATIONALE>
          Structured rationale for the act (authorship; requires --confidence)

      --persona <PERSONA>
          Persona/role the author acted as (authorship; requires --confidence)

      --model <MODEL>
          Model that authored the act (authorship; requires --confidence)

  -h, --help
          Print help (see a summary with '-h')

`temper resource facets`

List the live facets of a resource — one row per assert, with weights.

Sends `GET /api/resources/{id}/facets`. This is the faithful view: `resource show` carries a facet inside `open_meta` collapsed to a single value, newest-wins, with the weight dropped and any sibling row hidden. `facet_set` appends rather than upserts, so a resource can legitimately carry several live facet rows — this is where you see all of them.

Not to be confused with the `facets` key in `resource list`'s response envelope, which is search-style aggregate counts over the listed set, not properties of a resource.

Usage: temper resource facets [OPTIONS] <REF>

Arguments:
  <REF>
          Resource ref: a UUID or the decorated `slug-<uuid>` form

Options:
      --vault <VAULT>
          Path to vault (overrides TEMPER_VAULT and auto-detection)

      --format <FORMAT>
          Output format: json | toon (default: toon on a TTY, json otherwise). Precedence: --format → TEMPER_FORMAT → cli.format config → TTY default

      --embed-threads <N>
          ONNX intra-op threads for embedding. `0` = let ONNX Runtime decide. Default: this machine's performance-core count (NOT its total core count — efficiency cores measurably slow the batch down). Precedence: --embed-threads → TEMPER_ONNX_INTRA_THREADS → detected → 1

      --color <COLOR>
          Color output: auto | always | never (default: auto). Precedence: --color → TEMPER_COLOR → cli.color config → NO_COLOR → auto

  -h, --help
          Print help (see a summary with '-h')

resource

temper resource#

temper resource create#

temper resource list#

temper resource describe-open-meta#

temper resource doc-types#

temper resource describe-type#

temper resource connections#

temper resource citation-audits#

temper resource show#

temper resource evidence#

temper resource read-block#

temper resource update#

temper resource annotate#

temper resource meta#

temper resource meta get#

temper resource meta set#

temper resource audit-citation#

temper resource delete#

temper resource reassign#

temper resource grant#

temper resource revoke#

temper resource facet#

temper resource facets#

`temper resource`

`temper resource create`

`temper resource list`

`temper resource describe-open-meta`

`temper resource doc-types`

`temper resource describe-type`

`temper resource connections`

`temper resource citation-audits`

`temper resource show`

`temper resource evidence`

`temper resource read-block`

`temper resource update`

`temper resource annotate`

`temper resource meta`

`temper resource meta get`

`temper resource meta set`

`temper resource audit-citation`

`temper resource delete`

`temper resource reassign`

`temper resource grant`

`temper resource revoke`

`temper resource facet`

`temper resource facets`