@c4a/context-cli 0.7.29 → 0.7.30

package/plugins/cursor/commands/c4a-context-inspect-search.md

@@ -27,9 +27,9 @@ registrations, snapshots, tasks, approvals or user checkouts.
 Resolve `CONTEXT_QUERY_SOURCE_MODE` from the host or Bot instance configuration
 before choosing the first retrieval path. Supported values are:
 
-- `repo-first` (default): inspect the configured knowledge workspace and
-  authorized repository sources first. Use an available package only when the
-  repository path is unavailable or leaves a material gap.
+- `repo-first` (default): search approved `knowledge/` in the configured knowledge
+  workspace first; trace authorized original sources only for material gaps.
+  Use an available package when approved knowledge is unavailable.
 - `package-first`: inspect the selected package first, then use its recorded
   workspace or repository sources for missing, stale or disputed evidence.
 - `dual`: inspect the package and repository paths concurrently, then reconcile
@@ -43,8 +43,10 @@ sources and identify only gaps that affect the answer.
 ## Search available material according to the configured mode
 
 Apply `CONTEXT_QUERY_SOURCE_MODE` before starting retrieval. In `repo-first`,
-start with the configured workspace and reusable source checkouts; do not fetch
-or install a package merely to begin the query. In `package-first`, start with
+start with approved knowledge, not original-code checkout preparation. With
+readable `knowledge/`, do not probe `dist/`, build inventories or installed
+packages: distribution output is not a query prerequisite. An explicit request
+to investigate delivered content is an exception. In `package-first`, start with
 the selected readable package. In `dual`, start both paths concurrently. Search
 each selected path as soon as it is readable; tool checks and authorized
 upgrades must not block independent reading. If a configured primary path is
@@ -64,9 +66,11 @@ missing, use the host's configured, authorized package retrieval mechanism while
 other reading continues. Only check or prepare tools needed for the next actual
 operation; installation failure does not block independent local retrieval.
 
-For gaps or latest-content questions, also search approved knowledge when a workspace is available:
-use `knowledge/structure.yaml` and relevant `knowledge/` paths, trying business
-terms, symbols and synonyms, then read matching sections in context. Delivered
+Search relevant `knowledge/` paths using business terms, symbols and synonyms,
+then read matching sections in context; use `knowledge/structure.yaml` when
+navigation or associations are needed, not as an exhaustive preflight. In
+package-first or dual mode, approved knowledge can resolve gaps or freshness
+questions in delivered material. Delivered
 content may lag approved knowledge; they are not independent corroboration.
 Candidates and temporary reports are not approved pages. A search miss does not
 prove a capability is absent. Do not build merely to answer a query; a separately
@@ -76,8 +80,11 @@ When package or approved knowledge leaves a gap, directly trace relevant sources
 without asking whether to deepen the query. Use available document, code-search
 or extraction tools within the authorized scope, recording source identity and
 version. Prepare sources while continuing other reading.
-One failed source blocks only dependent claims. Answer when evidence is
-sufficient, without completing unnecessary preparation.
+One failed source blocks only dependent claims. If the read knowledge supports
+the requested conclusion, its conditions and scope, answer without restoring
+code or checking every associated source. Missing mechanisms, conflicting facts
+or unsupported current-behavior claims still require targeted investigation;
+an article title, search snippet or unread citation is not sufficient evidence.
 
 ## Trace and retrieve only relevant sources
 
@@ -88,18 +95,23 @@ registry is not a mandatory intermediate step. If the source identity or version
 cannot be established, retain package-grounded findings and qualify attribution;
 do not invent a repository or equate current source with the package's baseline.
 
-When a workspace and `context-build-inventory.json` are available, map a delivered page's
+Only when tracing a delivered page and a build inventory is already available, map its
 `dist_path` through `approved_knowledge.files` to its `approved_path`, relative
-to the workspace's `knowledge/` root. Follow the approved page's associations in
-`knowledge/structure.yaml` and `sources/*/index.yaml`. Without a reliable mapping,
+to the workspace's `knowledge/` root. For directly read approved pages, skip this
+mapping entirely. Follow relevant associations in `knowledge/structure.yaml`
+and `sources/*/index.yaml` only when source tracing is needed. Without a reliable mapping,
 continue package retrieval and follow explicit source references where available.
 Do not guess originals from similar filenames or claim attribution without
 checking the referenced material.
 
 Use the repository and recorded commit identified by the package or workspace
 as the source baseline.
-Reuse existing checkouts only after checking remote, commit, module coverage
-and local changes; do not reset them or change their sparse configuration.
+Before reusing a checkout for source reads, group checks of remote, commit,
+module coverage and local changes. Reuse those findings within this response
+instead of checking again before each search or file read. Recheck affected
+facts if the checkout/ref changes, sparse scope expands or there is evidence of
+concurrent edits; do not treat a previous response's checks as current. Do not
+reset user checkouts or change their sparse configuration.
 Missing code goes into a reusable query-owned path such as
 `.tmp/context-inspect/<host>/<repository-path>/<commit>/`, separate from user
 checkouts and production-managed sources. Use safe, credential-free path
@@ -128,10 +140,24 @@ capture/recovery actions. Production recovery gates do not block this independen
 inspection. Preserve unavailable-evidence gaps rather than changing production
 state to get past them.
 
-Compare relevant baseline files with current code when accessible, including
-relevant local changes. Use actual diffs to assess impact; current code is not a
-substitute for the baseline. Mention a missing comparison only when it limits
-the answer. For documents, notes and sessions, read saved bodies and necessary
+Batch independent searches and related file reads once paths are known, such as
+the entrypoint, state update and controlling render logic for one mechanism.
+Read enough surrounding context and follow unresolved imports or calls; batching
+must not truncate decisive evidence or turn a partial read into a complete audit.
+Reuse already read evidence rather than rereading it to prepare each tool call.
+
+Let evidence sufficiency determine whether a version comparison is needed.
+When the available knowledge or identified baseline code resolves the question,
+answer without fetching the default branch, recovering a second checkout or
+reading history merely to confirm freshness. Compare narrowly when uncertainty,
+conflicting evidence, suspected changes or an explicit version-comparison request
+makes it useful. A question about current behavior requires judging whether the
+available evidence supports that claim, not automatically comparing every branch.
+Retain necessary source reads; do not equate a recorded baseline or branch HEAD
+with a verified live deployment. When comparing, use actual relevant diffs,
+including relevant local changes; current code is not a substitute for the baseline.
+Mention a missing comparison only when it limits the answer.
+For documents, notes and sessions, read saved bodies and necessary
 attachments; fetch missing accessible material within scope. A summary is not a
 full transcript, an unread link is not evidence, and current remote content does
 not prove a historical snapshot.
@@ -164,13 +190,25 @@ limitation. Do not dump internal reasoning, runtime identifiers or routine
 version comparisons. Explain only differences affecting the answer; hashes may
 appear in source URLs without requiring a separate version audit in the prose.
 
-For website links, use `context-site-map.json` from the selected package or
-sibling website output. Match `pages[].package_path` or `approved_path`, and
+Collect citation targets after the relevant evidence is established and resolve
+them in one batch using the host's existing link resolver when available. Reuse
+known workspace, repository and site parameters; do not rediscover them per link.
+If later evidence adds targets, resolve only those missing from the results.
+Without a host resolver, use an already available `context-site-map.json` from
+the selected package or website output. Do not search for `dist/` merely to
+obtain links for readable approved knowledge. Match `pages[].package_path` or `approved_path`, and
 resolve `site_path` against `site_url` without prepending `base` again. Cite only
 matched pages, deduplicate links and do not invent anchors. If mapping is absent
 or invalid, retain local citations. Do not build, publish or probe remote sites
 just to format citations; a local map is not proof of the current online content.
 
+Once the requested questions are supported or genuinely unavailable evidence is
+clearly bounded, synthesize the final answer directly. Do not add another planning,
+freshness-audit or summary round just to prepare it. Preserve necessary mechanisms,
+exceptions, citations and verification limits; fewer preparation steps must not
+mean weaker evidence or omitted questions. Host delivery and receipt requirements
+still apply, without repeating the completed investigation.
+
 Suggest a knowledge update only with supporting evidence: identify the affected
 page or gap, proposed change and bounded source scope. Check neighboring content
 before proposing a new page. Only after user acceptance, hand the evidence and

package/plugins/cursor/skills/context-inspect-search/SKILL.md

@@ -28,9 +28,9 @@ registrations, snapshots, tasks, approvals or user checkouts.
 Resolve `CONTEXT_QUERY_SOURCE_MODE` from the host or Bot instance configuration
 before choosing the first retrieval path. Supported values are:
 
-- `repo-first` (default): inspect the configured knowledge workspace and
-  authorized repository sources first. Use an available package only when the
-  repository path is unavailable or leaves a material gap.
+- `repo-first` (default): search approved `knowledge/` in the configured knowledge
+  workspace first; trace authorized original sources only for material gaps.
+  Use an available package when approved knowledge is unavailable.
 - `package-first`: inspect the selected package first, then use its recorded
   workspace or repository sources for missing, stale or disputed evidence.
 - `dual`: inspect the package and repository paths concurrently, then reconcile
@@ -44,8 +44,10 @@ sources and identify only gaps that affect the answer.
 ## Search available material according to the configured mode
 
 Apply `CONTEXT_QUERY_SOURCE_MODE` before starting retrieval. In `repo-first`,
-start with the configured workspace and reusable source checkouts; do not fetch
-or install a package merely to begin the query. In `package-first`, start with
+start with approved knowledge, not original-code checkout preparation. With
+readable `knowledge/`, do not probe `dist/`, build inventories or installed
+packages: distribution output is not a query prerequisite. An explicit request
+to investigate delivered content is an exception. In `package-first`, start with
 the selected readable package. In `dual`, start both paths concurrently. Search
 each selected path as soon as it is readable; tool checks and authorized
 upgrades must not block independent reading. If a configured primary path is
@@ -65,9 +67,11 @@ missing, use the host's configured, authorized package retrieval mechanism while
 other reading continues. Only check or prepare tools needed for the next actual
 operation; installation failure does not block independent local retrieval.
 
-For gaps or latest-content questions, also search approved knowledge when a workspace is available:
-use `knowledge/structure.yaml` and relevant `knowledge/` paths, trying business
-terms, symbols and synonyms, then read matching sections in context. Delivered
+Search relevant `knowledge/` paths using business terms, symbols and synonyms,
+then read matching sections in context; use `knowledge/structure.yaml` when
+navigation or associations are needed, not as an exhaustive preflight. In
+package-first or dual mode, approved knowledge can resolve gaps or freshness
+questions in delivered material. Delivered
 content may lag approved knowledge; they are not independent corroboration.
 Candidates and temporary reports are not approved pages. A search miss does not
 prove a capability is absent. Do not build merely to answer a query; a separately
@@ -77,8 +81,11 @@ When package or approved knowledge leaves a gap, directly trace relevant sources
 without asking whether to deepen the query. Use available document, code-search
 or extraction tools within the authorized scope, recording source identity and
 version. Prepare sources while continuing other reading.
-One failed source blocks only dependent claims. Answer when evidence is
-sufficient, without completing unnecessary preparation.
+One failed source blocks only dependent claims. If the read knowledge supports
+the requested conclusion, its conditions and scope, answer without restoring
+code or checking every associated source. Missing mechanisms, conflicting facts
+or unsupported current-behavior claims still require targeted investigation;
+an article title, search snippet or unread citation is not sufficient evidence.
 
 ## Trace and retrieve only relevant sources
 
@@ -89,18 +96,23 @@ registry is not a mandatory intermediate step. If the source identity or version
 cannot be established, retain package-grounded findings and qualify attribution;
 do not invent a repository or equate current source with the package's baseline.
 
-When a workspace and `context-build-inventory.json` are available, map a delivered page's
+Only when tracing a delivered page and a build inventory is already available, map its
 `dist_path` through `approved_knowledge.files` to its `approved_path`, relative
-to the workspace's `knowledge/` root. Follow the approved page's associations in
-`knowledge/structure.yaml` and `sources/*/index.yaml`. Without a reliable mapping,
+to the workspace's `knowledge/` root. For directly read approved pages, skip this
+mapping entirely. Follow relevant associations in `knowledge/structure.yaml`
+and `sources/*/index.yaml` only when source tracing is needed. Without a reliable mapping,
 continue package retrieval and follow explicit source references where available.
 Do not guess originals from similar filenames or claim attribution without
 checking the referenced material.
 
 Use the repository and recorded commit identified by the package or workspace
 as the source baseline.
-Reuse existing checkouts only after checking remote, commit, module coverage
-and local changes; do not reset them or change their sparse configuration.
+Before reusing a checkout for source reads, group checks of remote, commit,
+module coverage and local changes. Reuse those findings within this response
+instead of checking again before each search or file read. Recheck affected
+facts if the checkout/ref changes, sparse scope expands or there is evidence of
+concurrent edits; do not treat a previous response's checks as current. Do not
+reset user checkouts or change their sparse configuration.
 Missing code goes into a reusable query-owned path such as
 `.tmp/context-inspect/<host>/<repository-path>/<commit>/`, separate from user
 checkouts and production-managed sources. Use safe, credential-free path
@@ -129,10 +141,24 @@ capture/recovery actions. Production recovery gates do not block this independen
 inspection. Preserve unavailable-evidence gaps rather than changing production
 state to get past them.
 
-Compare relevant baseline files with current code when accessible, including
-relevant local changes. Use actual diffs to assess impact; current code is not a
-substitute for the baseline. Mention a missing comparison only when it limits
-the answer. For documents, notes and sessions, read saved bodies and necessary
+Batch independent searches and related file reads once paths are known, such as
+the entrypoint, state update and controlling render logic for one mechanism.
+Read enough surrounding context and follow unresolved imports or calls; batching
+must not truncate decisive evidence or turn a partial read into a complete audit.
+Reuse already read evidence rather than rereading it to prepare each tool call.
+
+Let evidence sufficiency determine whether a version comparison is needed.
+When the available knowledge or identified baseline code resolves the question,
+answer without fetching the default branch, recovering a second checkout or
+reading history merely to confirm freshness. Compare narrowly when uncertainty,
+conflicting evidence, suspected changes or an explicit version-comparison request
+makes it useful. A question about current behavior requires judging whether the
+available evidence supports that claim, not automatically comparing every branch.
+Retain necessary source reads; do not equate a recorded baseline or branch HEAD
+with a verified live deployment. When comparing, use actual relevant diffs,
+including relevant local changes; current code is not a substitute for the baseline.
+Mention a missing comparison only when it limits the answer.
+For documents, notes and sessions, read saved bodies and necessary
 attachments; fetch missing accessible material within scope. A summary is not a
 full transcript, an unread link is not evidence, and current remote content does
 not prove a historical snapshot.
@@ -165,13 +191,25 @@ limitation. Do not dump internal reasoning, runtime identifiers or routine
 version comparisons. Explain only differences affecting the answer; hashes may
 appear in source URLs without requiring a separate version audit in the prose.
 
-For website links, use `context-site-map.json` from the selected package or
-sibling website output. Match `pages[].package_path` or `approved_path`, and
+Collect citation targets after the relevant evidence is established and resolve
+them in one batch using the host's existing link resolver when available. Reuse
+known workspace, repository and site parameters; do not rediscover them per link.
+If later evidence adds targets, resolve only those missing from the results.
+Without a host resolver, use an already available `context-site-map.json` from
+the selected package or website output. Do not search for `dist/` merely to
+obtain links for readable approved knowledge. Match `pages[].package_path` or `approved_path`, and
 resolve `site_path` against `site_url` without prepending `base` again. Cite only
 matched pages, deduplicate links and do not invent anchors. If mapping is absent
 or invalid, retain local citations. Do not build, publish or probe remote sites
 just to format citations; a local map is not proof of the current online content.
 
+Once the requested questions are supported or genuinely unavailable evidence is
+clearly bounded, synthesize the final answer directly. Do not add another planning,
+freshness-audit or summary round just to prepare it. Preserve necessary mechanisms,
+exceptions, citations and verification limits; fewer preparation steps must not
+mean weaker evidence or omitted questions. Host delivery and receipt requirements
+still apply, without repeating the completed investigation.
+
 Suggest a knowledge update only with supporting evidence: identify the affected
 page or gap, proposed change and bounded source scope. Check neighboring content
 before proposing a new page. Only after user acceptance, hand the evidence and

package/plugins/skills/context-inspect-search/SKILL.md

@@ -28,9 +28,9 @@ registrations, snapshots, tasks, approvals or user checkouts.
 Resolve `CONTEXT_QUERY_SOURCE_MODE` from the host or Bot instance configuration
 before choosing the first retrieval path. Supported values are:
 
-- `repo-first` (default): inspect the configured knowledge workspace and
-  authorized repository sources first. Use an available package only when the
-  repository path is unavailable or leaves a material gap.
+- `repo-first` (default): search approved `knowledge/` in the configured knowledge
+  workspace first; trace authorized original sources only for material gaps.
+  Use an available package when approved knowledge is unavailable.
 - `package-first`: inspect the selected package first, then use its recorded
   workspace or repository sources for missing, stale or disputed evidence.
 - `dual`: inspect the package and repository paths concurrently, then reconcile
@@ -44,8 +44,10 @@ sources and identify only gaps that affect the answer.
 ## Search available material according to the configured mode
 
 Apply `CONTEXT_QUERY_SOURCE_MODE` before starting retrieval. In `repo-first`,
-start with the configured workspace and reusable source checkouts; do not fetch
-or install a package merely to begin the query. In `package-first`, start with
+start with approved knowledge, not original-code checkout preparation. With
+readable `knowledge/`, do not probe `dist/`, build inventories or installed
+packages: distribution output is not a query prerequisite. An explicit request
+to investigate delivered content is an exception. In `package-first`, start with
 the selected readable package. In `dual`, start both paths concurrently. Search
 each selected path as soon as it is readable; tool checks and authorized
 upgrades must not block independent reading. If a configured primary path is
@@ -65,9 +67,11 @@ missing, use the host's configured, authorized package retrieval mechanism while
 other reading continues. Only check or prepare tools needed for the next actual
 operation; installation failure does not block independent local retrieval.
 
-For gaps or latest-content questions, also search approved knowledge when a workspace is available:
-use `knowledge/structure.yaml` and relevant `knowledge/` paths, trying business
-terms, symbols and synonyms, then read matching sections in context. Delivered
+Search relevant `knowledge/` paths using business terms, symbols and synonyms,
+then read matching sections in context; use `knowledge/structure.yaml` when
+navigation or associations are needed, not as an exhaustive preflight. In
+package-first or dual mode, approved knowledge can resolve gaps or freshness
+questions in delivered material. Delivered
 content may lag approved knowledge; they are not independent corroboration.
 Candidates and temporary reports are not approved pages. A search miss does not
 prove a capability is absent. Do not build merely to answer a query; a separately
@@ -77,8 +81,11 @@ When package or approved knowledge leaves a gap, directly trace relevant sources
 without asking whether to deepen the query. Use available document, code-search
 or extraction tools within the authorized scope, recording source identity and
 version. Prepare sources while continuing other reading.
-One failed source blocks only dependent claims. Answer when evidence is
-sufficient, without completing unnecessary preparation.
+One failed source blocks only dependent claims. If the read knowledge supports
+the requested conclusion, its conditions and scope, answer without restoring
+code or checking every associated source. Missing mechanisms, conflicting facts
+or unsupported current-behavior claims still require targeted investigation;
+an article title, search snippet or unread citation is not sufficient evidence.
 
 ## Trace and retrieve only relevant sources
 
@@ -89,18 +96,23 @@ registry is not a mandatory intermediate step. If the source identity or version
 cannot be established, retain package-grounded findings and qualify attribution;
 do not invent a repository or equate current source with the package's baseline.
 
-When a workspace and `context-build-inventory.json` are available, map a delivered page's
+Only when tracing a delivered page and a build inventory is already available, map its
 `dist_path` through `approved_knowledge.files` to its `approved_path`, relative
-to the workspace's `knowledge/` root. Follow the approved page's associations in
-`knowledge/structure.yaml` and `sources/*/index.yaml`. Without a reliable mapping,
+to the workspace's `knowledge/` root. For directly read approved pages, skip this
+mapping entirely. Follow relevant associations in `knowledge/structure.yaml`
+and `sources/*/index.yaml` only when source tracing is needed. Without a reliable mapping,
 continue package retrieval and follow explicit source references where available.
 Do not guess originals from similar filenames or claim attribution without
 checking the referenced material.
 
 Use the repository and recorded commit identified by the package or workspace
 as the source baseline.
-Reuse existing checkouts only after checking remote, commit, module coverage
-and local changes; do not reset them or change their sparse configuration.
+Before reusing a checkout for source reads, group checks of remote, commit,
+module coverage and local changes. Reuse those findings within this response
+instead of checking again before each search or file read. Recheck affected
+facts if the checkout/ref changes, sparse scope expands or there is evidence of
+concurrent edits; do not treat a previous response's checks as current. Do not
+reset user checkouts or change their sparse configuration.
 Missing code goes into a reusable query-owned path such as
 `.tmp/context-inspect/<host>/<repository-path>/<commit>/`, separate from user
 checkouts and production-managed sources. Use safe, credential-free path
@@ -129,10 +141,24 @@ capture/recovery actions. Production recovery gates do not block this independen
 inspection. Preserve unavailable-evidence gaps rather than changing production
 state to get past them.
 
-Compare relevant baseline files with current code when accessible, including
-relevant local changes. Use actual diffs to assess impact; current code is not a
-substitute for the baseline. Mention a missing comparison only when it limits
-the answer. For documents, notes and sessions, read saved bodies and necessary
+Batch independent searches and related file reads once paths are known, such as
+the entrypoint, state update and controlling render logic for one mechanism.
+Read enough surrounding context and follow unresolved imports or calls; batching
+must not truncate decisive evidence or turn a partial read into a complete audit.
+Reuse already read evidence rather than rereading it to prepare each tool call.
+
+Let evidence sufficiency determine whether a version comparison is needed.
+When the available knowledge or identified baseline code resolves the question,
+answer without fetching the default branch, recovering a second checkout or
+reading history merely to confirm freshness. Compare narrowly when uncertainty,
+conflicting evidence, suspected changes or an explicit version-comparison request
+makes it useful. A question about current behavior requires judging whether the
+available evidence supports that claim, not automatically comparing every branch.
+Retain necessary source reads; do not equate a recorded baseline or branch HEAD
+with a verified live deployment. When comparing, use actual relevant diffs,
+including relevant local changes; current code is not a substitute for the baseline.
+Mention a missing comparison only when it limits the answer.
+For documents, notes and sessions, read saved bodies and necessary
 attachments; fetch missing accessible material within scope. A summary is not a
 full transcript, an unread link is not evidence, and current remote content does
 not prove a historical snapshot.
@@ -165,13 +191,25 @@ limitation. Do not dump internal reasoning, runtime identifiers or routine
 version comparisons. Explain only differences affecting the answer; hashes may
 appear in source URLs without requiring a separate version audit in the prose.
 
-For website links, use `context-site-map.json` from the selected package or
-sibling website output. Match `pages[].package_path` or `approved_path`, and
+Collect citation targets after the relevant evidence is established and resolve
+them in one batch using the host's existing link resolver when available. Reuse
+known workspace, repository and site parameters; do not rediscover them per link.
+If later evidence adds targets, resolve only those missing from the results.
+Without a host resolver, use an already available `context-site-map.json` from
+the selected package or website output. Do not search for `dist/` merely to
+obtain links for readable approved knowledge. Match `pages[].package_path` or `approved_path`, and
 resolve `site_path` against `site_url` without prepending `base` again. Cite only
 matched pages, deduplicate links and do not invent anchors. If mapping is absent
 or invalid, retain local citations. Do not build, publish or probe remote sites
 just to format citations; a local map is not proof of the current online content.
 
+Once the requested questions are supported or genuinely unavailable evidence is
+clearly bounded, synthesize the final answer directly. Do not add another planning,
+freshness-audit or summary round just to prepare it. Preserve necessary mechanisms,
+exceptions, citations and verification limits; fewer preparation steps must not
+mean weaker evidence or omitted questions. Host delivery and receipt requirements
+still apply, without repeating the completed investigation.
+
 Suggest a knowledge update only with supporting evidence: identify the affected
 page or gap, proposed change and bounded source scope. Check neighboring content
 before proposing a new page. Only after user acceptance, hand the evidence and

package/providers/context/manifest.json

@@ -2,7 +2,7 @@
   "schema": "agent-graph.bundle.v1",
   "provider": {
     "id": "c4a/context",
-    "version": "0.7.29"
+    "version": "0.7.30"
   },
   "providerManifest": "provider.yaml",
   "graphs": [
@@ -183,7 +183,7 @@
     {
       "id": "context.sdk.lark-resources",
       "path": "resources/manuals/guides/lark-resources.md",
-      "digest": "sha256:b1ffb74a145962ff89a5db29cffd1318e4ec038b97884dcd43804170cb0f6f64"
+      "digest": "sha256:d5b6b44fe800f302232300adbdb4f32b04b78588ea7879dc9a5c247f36658b9d"
     },
     {
       "id": "context.sdk.package-outputs",
@@ -248,7 +248,7 @@
     {
       "id": "dialogue.document-capture",
       "path": "resources/dialogue/document-capture.md",
-      "digest": "sha256:eab34072dc96d77c0749148463f722d2483d8e7d83899ce489e927176a1178e4"
+      "digest": "sha256:20bb708f2c471591c0f633742d75a1353bcd4762917ce9528d3de0c7f4babbe8"
     },
     {
       "id": "dialogue.human-gates",
@@ -288,7 +288,7 @@
     {
       "id": "procedure.document-capture",
       "path": "resources/procedures/document-capture.md",
-      "digest": "sha256:9d0d1b9187bbe540a4afffb366cd549385dbf61bfaffa96ae51de848b2dd5460"
+      "digest": "sha256:2e6b4e45587fffc1930a3dce0fc6b68d659499a034a367e1cc34a2ef46cb50d9"
     },
     {
       "id": "procedure.document-code-investigation",
@@ -368,7 +368,7 @@
     {
       "id": "procedure.source-capture-detailed",
       "path": "resources/procedures/source-capture-detailed.md",
-      "digest": "sha256:75694361543f5c44a1077b5e6c4ed9bca844ec9b46c1773198f66e0f20ba3748"
+      "digest": "sha256:39644875f09bac66580332357c3d9701c234da036b61205c8a02029165455f76"
     },
     {
       "id": "procedure.source-update",
@@ -575,7 +575,7 @@
     },
     {
       "path": "provider.yaml",
-      "digest": "sha256:0a21f82483186c4fe1f6f450923d0d336300274fb95e2c7ee3569700ba922372"
+      "digest": "sha256:50adfb493675f902e6df4d6eade84bb279c64b767084aa15277d65870825458e"
     },
     {
       "path": "resources/contracts/indexer-provider-guide.yaml",
@@ -599,7 +599,7 @@
     },
     {
       "path": "resources/dialogue/document-capture.md",
-      "digest": "sha256:eab34072dc96d77c0749148463f722d2483d8e7d83899ce489e927176a1178e4"
+      "digest": "sha256:20bb708f2c471591c0f633742d75a1353bcd4762917ce9528d3de0c7f4babbe8"
     },
     {
       "path": "resources/dialogue/human-gates.md",
@@ -635,7 +635,7 @@
     },
     {
       "path": "resources/manuals/guides/lark-resources.md",
-      "digest": "sha256:b1ffb74a145962ff89a5db29cffd1318e4ec038b97884dcd43804170cb0f6f64"
+      "digest": "sha256:d5b6b44fe800f302232300adbdb4f32b04b78588ea7879dc9a5c247f36658b9d"
     },
     {
       "path": "resources/manuals/guides/markdown-indexer-skill-authoring.md",
@@ -695,7 +695,7 @@
     },
     {
       "path": "resources/procedures/document-capture.md",
-      "digest": "sha256:9d0d1b9187bbe540a4afffb366cd549385dbf61bfaffa96ae51de848b2dd5460"
+      "digest": "sha256:2e6b4e45587fffc1930a3dce0fc6b68d659499a034a367e1cc34a2ef46cb50d9"
     },
     {
       "path": "resources/procedures/document-code-investigation.md",
@@ -759,7 +759,7 @@
     },
     {
       "path": "resources/procedures/source-capture-detailed.md",
-      "digest": "sha256:75694361543f5c44a1077b5e6c4ed9bca844ec9b46c1773198f66e0f20ba3748"
+      "digest": "sha256:39644875f09bac66580332357c3d9701c234da036b61205c8a02029165455f76"
     },
     {
       "path": "resources/procedures/source-update.md",
@@ -878,5 +878,5 @@
     "indexer": [],
     "workspace": []
   },
-  "digest": "sha256:3e0f9fc6f81571381dced3a8d02b20624022183e924c7b99c16430c8924d624d"
+  "digest": "sha256:36c06f801f991b92bb78cc94676c8d338389bdc031f49577eb196a96628b2f76"
 }

package/providers/context/provider.yaml

@@ -1,6 +1,6 @@
 schema: agent-graph.provider.v1
 id: c4a/context
-version: 0.7.29
+version: 0.7.30
 name: Context workflow
 description: Internal work contract for Context knowledge workspaces.
 graphs:

package/providers/context/resources/dialogue/document-capture.md

@@ -23,6 +23,14 @@ refresh, or capture preview before permission is clear. A request that
 explicitly says to ingest, fetch, capture, or read the named documents grants
 that scope; merely mentioning a possible source does not.
 
+The image preflight in the document-capture procedure is an Agent self-check,
+not an additional user question. Before capture, briefly report the task-wide
+document count and applied policy. Reference-only images keep source references
+but are not downloaded or interpreted. Do not wait for confirmation or infer an
+all-image override from managed execution. Unknown scope requires authorized
+metadata or body-and-reference-only discovery first. Existing source-read
+permission requirements remain unchanged.
+
 If deterministic inspection detects `.mdx`, route metadata, sidebars, or a docs
 configuration, explain that the boundary looks like a documentation site and
 ask whether to use the Context document-site processor. Do not infer that

package/providers/context/resources/manuals/guides/lark-resources.md

@@ -133,8 +133,24 @@ assess the page with its non-text evidence.
 
 ## Image choices for a production task
 
+Before capture, the Agent follows the document-capture Route procedure's task-wide
+document threshold and image self-check, without asking for user confirmation.
+Set `DOCUMENT_IMAGE_CAPTURE` in task instructions or workspace guidance to `ON`
+(include images), `OFF` (references only), or a positive integer document threshold.
+Omitting it uses `10`; exactly the threshold remains within normal acquisition.
+Current-conversation instructions override workspace guidance for that task only.
+For example, `DOCUMENT_IMAGE_CAPTURE: 20` skips image downloads when the task
+contains more than 20 distinct documents. This is an Agent instruction, not an
+environment variable, CLI flag or SDK field; no Bot configuration is required.
+Above a numeric threshold, the default is body capture with image references,
+unless the effective setting is overridden to `ON`. This workflow
+choice uses the existing resource fields; it does not change SDK defaults or add
+a CLI hard gate. Apply it before bulk capture or host prefetch, not after images
+have already been downloaded.
+
 The work-start report asks once when a task contains more than 30 distinct images
-and has no explicit image policy. The choices are intelligent conversion and
+that will be acquired and has no explicit image handling policy. Reuse the
+current-conversation image instruction when it already settles this choice. The choices are intelligent conversion and
 inclusion (recommended), include all as images with compression, or include none
 with placeholders. Count the whole task, not each capture or writing batch.
 Review reports show the resulting image handling and any fallbacks.