fix(agent): fit fork_context into the existing tool-schema budget

Advertising `fork_context` on the agent tool grew every mode's tool
schema by 253 bytes and failed the runtime-contract ratchet. Instead of
raising the ceiling, drop the no-op `fork_context: {}` from the start
dependentSchema branch and tighten the `resume_from`, `fork_context` and
`wall_time_secs` descriptions (same facts, fewer words). Net: -14 bytes
per mode versus main, so the budget is tightened, and
PARENT_SURFACE_BYTE_CEILING returns to main's 88,715.

Local checks:
- check-runtime-contract-budget.py: PASS, 12 ceilings tightened (--update)
- codewhale-tui --lib tools::subagent: 738 passed, 0 failed
- codewhale-tui --lib tool_catalog/schema: 166 passed, 0 failed, 1 ignored
- test_check_runtime_contract_budget.py OK; blocking-calls budget within
- check-versions.sh --range-audit-advisory OK
- cargo fmt --all, git diff --check clean

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01R9rJEoMuUSRWznU6QjkE7h

crates/tui/src/tools/subagent/mod.rs

@@ -10134,7 +10134,7 @@ impl ToolSpec for AgentTool {
                 },
                 "wall_time_secs": {
                     "type": "integer", "minimum": 1, "maximum": MAX_CHILD_WALL_TIME.as_secs(),
-                    "description": "Whole-run wall time including queue, model and tools; default 1800. May raise that default, never inherited or operator limits; continuation does not restart the clock."
+                    "description": "Whole-run wall time (queue, model, tools); default 1800, may exceed that but never inherited or operator limits. Continuation keeps the clock."
                 },
                 "deliverables": {
                     "type": "array", "maxItems": 16,
@@ -10159,11 +10159,11 @@ impl ToolSpec for AgentTool {
                 },
                 "fork_context": {
                     "type": "boolean",
-                    "description": "For start: true = child starts from your current conversation prefix (cache-shared); false (default) = fresh context. Carries context, never permissions."
+                    "description": "For start: true = child starts from your conversation prefix (cache-shared); default fresh. Never grants permissions."
                 },
                 "resume_from": {
                     "type": "string",
-                    "description": "Settled child agent_id or session name to fork into a separate new worker. Repeating start with resume_from creates another independent worker; use action=followup to continue parked work without an accidental duplicate. The source must not be running. Its full transcript is loaded and prepended as the new child's context (fork_context=true), continuing the transcript lineage under a new role or profile (e.g. explore → implementer → verifier). Mutually exclusive with fork_context=false. Cross-workspace or missing sources are rejected with a clear error."
+                    "description": "Settled child agent_id or session name to fork into a new independent worker; each start makes another, so use action=followup to continue parked work. Its full transcript becomes the new child's context, continuing the lineage under a new role or profile (e.g. explore → implementer → verifier). Refused with fork_context=false, or for cross-workspace or missing sources."
                 }
             },
             "dependentSchemas": {
@@ -10172,8 +10172,7 @@ impl ToolSpec for AgentTool {
                         {
                             "properties": {
                                 "action": {"const": "start"},
-                                "prompt": {},
-                                "fork_context": {}
+                                "prompt": {}
                             },
                             "required": ["prompt"]
                         },

crates/tui/src/tools/subagent/tests.rs

@@ -21187,11 +21187,9 @@ const READ_ONLY_CHILD_ENVELOPE_BYTE_CEILING: usize = 89_000;
 // base prompt's progress-narration rule (E4, 5cf9db3d6) and the workflow
 // Fleet origin list (26cfaf8de), net of the read/bash wording trims
 // (105ad9d3e).
-// Re-measured 2026-09-25 at 88,937B on Linux (88,924B on macOS), +222B: the
-// agent schema advertises `fork_context` (it was parse-accepted but hidden).
-// Re-measured 2026-09-25 at 88,968B on Linux (88,955B on macOS), +31B: the
-// `wall_time_secs` description states that it may raise the 1800s default.
-const PARENT_SURFACE_BYTE_CEILING: usize = 88_968;
+// The agent schema's `fork_context` property (2026-09-25) fit under this
+// ceiling by trimming the `resume_from` and `wall_time_secs` descriptions.
+const PARENT_SURFACE_BYTE_CEILING: usize = 88_715;
 
 #[tokio::test]
 async fn read_only_child_envelope_stays_within_measured_ceiling() {

scripts/runtime-contract-budget.json

@@ -86,9 +86,9 @@
     "modes": {
       "act": {
         "active": {
-          "bytes": 33876,
+          "bytes": 33862,
           "identity_sha256": "df6676989a677fb08fecc8bf7ae12caf4fa88e0cae3d143cea6a4da4382b746c",
-          "tokens_est": 8469,
+          "tokens_est": 8466,
           "tool_names": [
             "agent",
             "bash",
@@ -106,9 +106,9 @@
           "tools": 12
         },
         "full": {
-          "bytes": 83384,
+          "bytes": 83370,
           "identity_sha256": "45e989bbe5ac0bb1f2d9084361c021009f30a06539f132ebd4fd2331a1bb1954",
-          "tokens_est": 20846,
+          "tokens_est": 20843,
           "tool_names": [
             "Git",
             "Run",
@@ -169,9 +169,9 @@
       },
       "operate": {
         "active": {
-          "bytes": 33876,
+          "bytes": 33862,
           "identity_sha256": "df6676989a677fb08fecc8bf7ae12caf4fa88e0cae3d143cea6a4da4382b746c",
-          "tokens_est": 8469,
+          "tokens_est": 8466,
           "tool_names": [
             "agent",
             "bash",
@@ -189,9 +189,9 @@
           "tools": 12
         },
         "full": {
-          "bytes": 83384,
+          "bytes": 83370,
           "identity_sha256": "45e989bbe5ac0bb1f2d9084361c021009f30a06539f132ebd4fd2331a1bb1954",
-          "tokens_est": 20846,
+          "tokens_est": 20843,
           "tool_names": [
             "Git",
             "Run",
@@ -252,9 +252,9 @@
       },
       "plan": {
         "active": {
-          "bytes": 33876,
+          "bytes": 33862,
           "identity_sha256": "df6676989a677fb08fecc8bf7ae12caf4fa88e0cae3d143cea6a4da4382b746c",
-          "tokens_est": 8469,
+          "tokens_est": 8466,
           "tool_names": [
             "agent",
             "bash",
@@ -272,9 +272,9 @@
           "tools": 12
         },
         "full": {
-          "bytes": 53515,
+          "bytes": 53501,
           "identity_sha256": "ac8af1f4988199825be7b00b054c258724a44074b1d4e6de6c92ade7c1cffe63",
-          "tokens_est": 13379,
+          "tokens_est": 13376,
           "tool_names": [
             "Git",
             "Web",