Follow-up to #958: the advisor-spend record's four P3s and the cross-client page's Claude sources

.claude/skills/claude-native-practice/reference/cross-client.md

@@ -31,3 +31,20 @@ Sources: <https://claude.com/blog/code-review>, <https://www.anthropic.com/resea
 | Packaged capability | plugins (skills, agents, hooks, MCP) | `codex plugin add` (skills, MCP) |
 | Native review | `/review-changes` (saved Workflow), `/code-review` in the main session | `codex review --base <ref>` or `--commit <sha>` |
 | Clean-context run for an eval | a fresh subagent or `claude -p` session | `codex exec --ephemeral` |
+
+Claude column sources: the Claude Code documentation pages the 2026-10-09 reading captured. For each citation, the URL, the capture's
+sha256 and the fetch time are locators of `evidence/artifacts/claude-native-practice-20261009/claims.json`, and the line (counted by newline
+in that capture) is recorded with the sha256 of its text in `evidence/artifacts/claude-native-practice-20261009/cross-client-lines.json`; a
+test checks the page against both. The captures are private working papers of the reading, not in this repository: with a capture,
+`sed -n '<line>p' <capture> | sha256sum` prints the recorded hash of a line that is followed by a newline (each cited line is). `/review-changes` is this repository's saved Workflow,
+`examples/claude-native/workflows/review-changes.js`.
+
+| Need | Claude Code source |
+| --- | --- |
+| Delegation with fresh context | [sub-agents](https://code.claude.com/docs/en/sub-agents.md) line 11, sha256 `b4e0ef9b2d9a1cba`, fetched 2026-10-09T03:39:03Z (each subagent runs in its own context window); [sub-agents](https://code.claude.com/docs/en/sub-agents.md) line 492, sha256 `b4e0ef9b2d9a1cba`, fetched 2026-10-09T03:39:03Z (the Agent tool, renamed from Task in 2.1.63); [workflows](https://code.claude.com/docs/en/workflows.md) line 204, sha256 `e55138abf9b78682`, fetched 2026-10-09T03:39:40Z (saved Workflows in the project's .claude/workflows/) |
+| Completion gate | [hooks](https://code.claude.com/docs/en/hooks.md) line 2587, sha256 `3bd21d31955304d9`, fetched 2026-10-09T03:38:41Z (the Stop input carries stop_hook_active); [hooks](https://code.claude.com/docs/en/hooks.md) line 2655, sha256 `3bd21d31955304d9`, fetched 2026-10-09T03:38:41Z (decision block prevents Claude from stopping) |
+| Structured headless result | [cli-reference](https://code.claude.com/docs/en/cli-reference.md) line 103, sha256 `54587c07d51da9f2`, fetched 2026-10-09T03:38:34Z (--json-schema, print mode only) |
+| Machine-readable event stream | [cli-reference](https://code.claude.com/docs/en/cli-reference.md) line 112, sha256 `54587c07d51da9f2`, fetched 2026-10-09T03:38:34Z (--output-format text, json or stream-json) |
+| Packaged capability | [plugins](https://code.claude.com/docs/en/plugins.md) line 9, sha256 `eb8b9f92d75d42d7`, fetched 2026-10-09T03:39:03Z (a plugin is a directory of skills, agents, hooks and MCP servers) |
+| Native review | [commands](https://code.claude.com/docs/en/commands.md) line 71, sha256 `98494561aac56c11`, fetched 2026-10-09T03:41:45Z (/code-review) |
+| Clean-context run for an eval | [headless](https://code.claude.com/docs/en/headless.md) line 17, sha256 `25885d2e178a9746`, fetched 2026-10-09T03:38:34Z (claude -p); [sub-agents](https://code.claude.com/docs/en/sub-agents.md) line 11, sha256 `b4e0ef9b2d9a1cba`, fetched 2026-10-09T03:39:03Z (a fresh subagent context) |

evidence/artifacts/claude-native-practice-20261009/cross-client-lines.json

@@ -0,0 +1,86 @@
+{
+ "schema": "claude-native-practice-cross-client-lines/1",
+ "note": "The lines the Claude source table of reference/cross-client.md cites, with the capture each was read from. Lines are counted by newline in the capture. line_sha256 is the sha256 of the line's bytes followed by a newline. For a line that is followed by a newline in the capture (each recorded line is), `sed -n '<line>p' <capture> | sha256sum` prints the same. The captures are private working papers of the 2026-10-09 reading, not in this repository; their hashes and fetch times are the locators of claims.json, which the guard test checks.",
+ "lines": [
+  {
+   "url": "https://code.claude.com/docs/en/cli-reference.md",
+   "line": 103,
+   "capture_sha256": "54587c07d51da9f23418857237434cd0140561736c943296e247d8e13cb85a58",
+   "fetched_utc": "2026-10-09T03:38:34Z",
+   "line_sha256": "efeb1a08813b39efcbb98709e8477a563805ec4ef94d5fdccfd9ba433e95b6ff",
+   "line_head": "| `--json-schema` | Get validated JSON output matching a JSON Schema after the a"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/cli-reference.md",
+   "line": 112,
+   "capture_sha256": "54587c07d51da9f23418857237434cd0140561736c943296e247d8e13cb85a58",
+   "fetched_utc": "2026-10-09T03:38:34Z",
+   "line_sha256": "d82d212656cfcd9c61c26d8a927ed7078ce99ca04d3e66949bd45d611253bf05",
+   "line_head": "| `--output-format` | Specify output format for print mode (options: `text`, `js"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/commands.md",
+   "line": 71,
+   "capture_sha256": "98494561aac56c1106a49955ba503529d4ae6a21ef1b593c187b4d22465caae0",
+   "fetched_utc": "2026-10-09T03:41:45Z",
+   "line_sha256": "58efa08a8511d6330546bdb3f35fff17778552e78c79187425563dbe0f7408f9",
+   "line_head": "| `/code-review [low\\|medium\\|high\\|xhigh\\|max\\|ultra] [--fix] [--comment] [--ma"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/headless.md",
+   "line": 17,
+   "capture_sha256": "25885d2e178a9746066fc2e19a07a3debc31744c721282a744be80789591e738",
+   "fetched_utc": "2026-10-09T03:38:34Z",
+   "line_sha256": "465e4fdd7bb425d042d9645b542d01eab7a97868b3704fbab38e63390277ba9e",
+   "line_head": "This page covers using the Agent SDK via the CLI (`claude -p`). For the Python a"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/hooks.md",
+   "line": 2587,
+   "capture_sha256": "3bd21d31955304d9ab33efb052a0b3bfdfeb19d1af5028a2588d69c5d9bdea95",
+   "fetched_utc": "2026-10-09T03:38:41Z",
+   "line_sha256": "6b32c97574311d127912ff1230241e94eb7c3c5457b22e54a1b91c4388255d77",
+   "line_head": "In addition to the [common input fields](#common-input-fields), Stop hooks recei"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/hooks.md",
+   "line": 2655,
+   "capture_sha256": "3bd21d31955304d9ab33efb052a0b3bfdfeb19d1af5028a2588d69c5d9bdea95",
+   "fetched_utc": "2026-10-09T03:38:41Z",
+   "line_sha256": "3d16e333ffa379ee122d7ad20c9b50b624b148806e9f0f07217c7103047993b7",
+   "line_head": "| `decision` | `\"block\"` prevents Claude from stopping. Omit to allow Claude to "
+  },
+  {
+   "url": "https://code.claude.com/docs/en/plugins.md",
+   "line": 9,
+   "capture_sha256": "eb8b9f92d75d42d777d3bdbc89273c2df881b6a3cf58afaa63418d55a70538f5",
+   "fetched_utc": "2026-10-09T03:39:03Z",
+   "line_sha256": "21fdcfc941dd9c2bab11f6e95d4024c0a257b48f535f87adae8f903acdebcf8c",
+   "line_head": "A Claude Code plugin is a directory of skills, agents, hooks, MCP servers, or ot"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/sub-agents.md",
+   "line": 11,
+   "capture_sha256": "b4e0ef9b2d9a1cba1a53c45a5d45b0c525d949261c8beb16d60dd35fbcb134d8",
+   "fetched_utc": "2026-10-09T03:39:03Z",
+   "line_sha256": "acae0fc77c92163c0a7e05e5a05ce7b1341f82363739e837a4093d9830face06",
+   "line_head": "Each subagent runs in its own context window with a custom system prompt, specif"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/sub-agents.md",
+   "line": 492,
+   "capture_sha256": "b4e0ef9b2d9a1cba1a53c45a5d45b0c525d949261c8beb16d60dd35fbcb134d8",
+   "fetched_utc": "2026-10-09T03:39:03Z",
+   "line_sha256": "a2e450073758fd52f939d2117015286beb3a8fd1bc4287506fb85a6874fe68c4",
+   "line_head": "<Note>In version 2.1.63, the Task tool was renamed to Agent. Existing `Task(...)"
+  },
+  {
+   "url": "https://code.claude.com/docs/en/workflows.md",
+   "line": 204,
+   "capture_sha256": "e55138abf9b78682085c09b1b7632254461169eeed738f8b2a532849498e0c8e",
+   "fetched_utc": "2026-10-09T03:39:40Z",
+   "line_sha256": "6b7f4fd880cd8e3c61b67510a5eaefb86fcbdff6a31dc1b8b250be819ccbe80c",
+   "line_head": "* `.claude/workflows/` in your project: shared with everyone who clones the repo"
+  }
+ ]
+}

evidence/artifacts/claude-native-practice-20261009/primary-source-reading.json

@@ -1,6 +1,6 @@
 {
  "schema": "claude-native-practice-primary-reading/1",
- "run": "wf_bee0b670-035 (2026-10-09 13:48-14:12Z): per source an Opus 5.5 xhigh reader extracted the code and workflow quality practices it states and mapped each to a role slot and client; an Opus 5.5 max refuter re-read the page and refuted each practice by default; kept = not refuted; the children could call the advisor tool, a Fable 5.1 advisor: 63 calls, 29 answered (8 by readers, 21 by refuters, in 28 children) and 34 refused as too_many_requests, all billed at the advisor's rates outside the executors' usage",
+ "run": "wf_bee0b670-035 (2026-10-09 13:48-14:12Z): per source an Opus 5.5 xhigh reader extracted the code and workflow quality practices it states and mapped each to a role slot and client; an Opus 5.5 max refuter re-read the page and refuted each practice by default; kept = not refuted; the children could call the advisor tool, a Fable 5.1 advisor: 63 calls, 29 answered (8 by readers, 21 by refuters, in 28 children) and 34 refused as too_many_requests; the answered calls bill at the advisor's rates outside the executors' usage, the refused calls carry no usage and add no advisor charge (the executor tokens that issue a refused call stay in the executors' spend)",
  "counts": {
   "sources": 31,
   "practices": 248,
@@ -76,7 +76,7 @@
   },
   "pricing": {
    "source": "https://platform.claude.com/docs/en/about-claude/pricing",
-   "read_utc": "2026-10-10T14:23Z",
+   "read_utc": "2026-10-10T15:24:32Z",
    "page_sha256": "67904e280b595d31f373c857844ed380072ae79d280d8f385640b5992d337ca9",
    "advisor_billing_source": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool (Usage and billing: advisor calls are billed at the advisor model's rates; the top-level usage fields reflect executor tokens only and do not roll the advisor's tokens in)",
    "advisor_page_sha256": "81a97b998b6d8043032a656214e34ffb474b3f4fa37d2724155df396ffefaf5a",
@@ -96,7 +96,20 @@
      "output": 20
     }
    },
-   "method": "tokens times the table's rate per million tokens, summed over the five token classes of each model and rounded to cents after summing; the executor's cache creation is all 5-minute (the record's per-message split)"
+   "method": "tokens times the table's rate per million tokens, summed over the five token classes of each model and rounded to cents after summing; the executor's cache creation is all 5-minute (the record's per-message split)",
+   "read_note": "both pages re-fetched by the script that wrote this field (p958_p3_fix.py, sha256 d21b6dc2e3b50f5a68e4a09c0540cf20a754f307c718cce53687cac7bcb8cf4b, kept in the research directory), as the .md files named in hashed_urls, at this time, and hashed to the values above (the HTML renderings of the same pages hash differently and change with the site); the first read, at about 14:23Z, gave the same hashes",
+   "assumptions": {
+    "speed": "standard list rates throughout. 677 executor messages record speed standard; 24 record no speed (service tier standard; 0.85 USD of the executors' spend at standard list) and the 29 advisor iterations record neither speed nor service tier, which the API does not report for an advisor iteration, so the figure assumes standard speed and routing for those",
+    "executor_messages_with_speed_standard": 677,
+    "executor_messages_without_speed": 24,
+    "executor_messages_without_speed_usd_at_standard_list": 0.85,
+    "advisor_iterations_without_speed_or_tier": 29,
+    "modifiers_not_applied": "none recorded: no fast mode and no web tools in the run; cache hits at 0.05x (Opus 5.5) and 0.025x (Fable 5.1) of the base input price are the table's own cache-hit column"
+   },
+   "hashed_urls": {
+    "page_sha256": "https://platform.claude.com/docs/en/about-claude/pricing.md",
+    "advisor_page_sha256": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool.md"
+   }
   },
   "priced_usd_executor_api_list": 63.87,
   "priced_usd_advisor_api_list": 27.64,

manifests/evidence.json

@@ -4941,8 +4941,8 @@
     },
     {
       "path": ".claude/skills/claude-native-practice/reference/cross-client.md",
-      "sha256": "24ce9c3d0b3ce48104176d3737a441402524a548fbd2c57af034e93e490460df",
-      "bytes": 2041
+      "sha256": "13f157f6d35e858bc1407664645e95c0a598b577c4d787a3da3687162e6b6570",
+      "bytes": 4875
     },
     {
       "path": ".claude/skills/claude-native-practice/reference/git-github-automation.md",
@@ -20030,6 +20030,11 @@
       "sha256": "1ada50b39afc9a02b29f5a28a935c7bcdf8615140d0aa7d0229d92cef8f85448",
       "bytes": 236851
     },
+    {
+      "path": "evidence/artifacts/claude-native-practice-20261009/cross-client-lines.json",
+      "sha256": "7874d2ee5c90bfb6d47f0f58ca29355696e0476ebd3a2a6c830d00e6ce2303d6",
+      "bytes": 4622
+    },
     {
       "path": "evidence/artifacts/claude-native-practice-20261009/fresh-session.json",
       "sha256": "5db3955c7b24d3202d9a43c6f002a194e6091b007d3d68435bde982d485e8363",
@@ -20047,8 +20052,8 @@
     },
     {
       "path": "evidence/artifacts/claude-native-practice-20261009/primary-source-reading.json",
-      "sha256": "0d6807c6a7dd4e51a46fffd641123a0203749db613bba4d30b60cc6ac1c40eea",
-      "bytes": 445981
+      "sha256": "1113e5cdefa53d9e18bbb6a9b7c568eedd5949d4d2a1e7bbc0e12696a3101b71",
+      "bytes": 447581
     },
     {
       "path": "evidence/artifacts/claude-native-practice-20261009/probes-20261009.json",
@@ -60795,6 +60800,11 @@
       "sha256": "8997b51d8400a77eb8a182254cd22e348a7cfde967b367373819d5c3af879eea",
       "bytes": 5702
     },
+    {
+      "path": "tests/test_native_practice_cross_client_sources.py",
+      "sha256": "6c7334aa820914a96d9d5b9228630ab1e1621e03ca6ece86bc76124f3fd9a3d3",
+      "bytes": 62924
+    },
     {
       "path": "tests/test_native_practice_doc_corrections.py",
       "sha256": "04410b557212f2cd75a0ac37a5d6fd0addd9920d7f8caaf8a057a8d0ba9432e1",
@@ -60982,8 +60992,8 @@
     },
     {
       "path": "tests/test_primary_source_reading_counts.py",
-      "sha256": "d5adf976f213cdff48d0c1ec245d58596203760d9f3ffb3253baedbfbc228b88",
-      "bytes": 20294
+      "sha256": "fe4e5e9125f0a3e6927ef1ab74ea62bbc16abbd64faf32aa88fe790b6e35041d",
+      "bytes": 24282
     },
     {
       "path": "tests/test_promotion_gate.py",