tradingcodex 2.1.1__tar.gz → 2.1.2__tar.gz

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/CHANGELOG.md

@@ -2,6 +2,18 @@
 
 ## Unreleased
 
+## 2.1.2 - 2026-07-24
+
+- Treat ASCII skill-name case and normalized whitespace layout as presentation
+  details, so Codex app skill-chip capitalization, Brain/Strategy selectors on
+  separate lines, and line-wrapped order arguments bind to their canonical
+  lowercase skills without weakening order flags, single-effect grants,
+  approvals, policy, idempotency, or audit enforcement.
+- Let evidence-producing subagents autonomously fill newly discovered material
+  gaps and conflicts within their assigned question, specialty, user scope, and
+  read-only authority, using evidence value instead of fixed search or tool-call
+  counts while preserving source provenance, deduplication, and role ownership.
+
 ## 2.1.1 - 2026-07-24
 
 - Reframe independent judgment review as task-relative adversarial adjudication:

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tradingcodex
-Version: 2.1.1
+Version: 2.1.2
 Summary: Local-first investment operating system for Codex-native research and service-gated execution.
 Author: TradingCodex Authors
 License-Expression: Apache-2.0

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/docs/generated-workspaces.md

@@ -323,11 +323,13 @@ Generated workspaces contain:
   reference a Head Manager may pass onward
 - deterministic root-native action interception: before allocating an analysis
   run, `UserPromptSubmit` reserves immediate `$tcx-order-submit` and
-  `$tcx-order-cancel`, accepts only their complete exact `--name value`
-  grammar from a root native user turn, creates a workspace-bound `native-user`
+  `$tcx-order-cancel`, accepts their complete literal `--name value`
+  grammar from a root native user turn while tolerating ASCII skill-name case
+  and whitespace line wrapping, creates a workspace-bound `native-user`
   mandate, and calls the canonical Django execution gateway in-process. It also
-  reserves a first-meaningful-line `$tcx-order-allow --mode
-  paper|validation|live`, issues one workspace/session/turn/prompt/mode-bound
+  reserves a first-meaningful `$tcx-order-allow` invocation followed by
+  `--mode paper|validation|live` and a non-empty request on the same or later
+  lines, issues one workspace/session/turn/prompt/mode-bound
   `OrderTurnGrant`, and then continues the normal workflow. Malformed reserved
   actions, subagent actions, and retired
   `$execute-paper-order` actions fail closed

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/docs/harness.md

@@ -67,10 +67,15 @@ adequate Snapshot/Dataset
   -> explicit gap
 ```
 
-Partial success is retained and only missing coverage falls through. External
-sources become SourceSnapshots; reusable structured rows become immutable
-Datasets. Handoffs carry compact Snapshot/Dataset/Artifact IDs rather than raw
-source dumps. This is skill guidance, not a provider router or trust engine.
+Partial success is retained and unsupported coverage falls through without
+refetching adequate evidence. Within its assigned question and specialty, an
+evidence-producing child may autonomously investigate a newly discovered
+material gap or conflict; no fixed source or tool-call count applies. Broad
+just-in-case collection and another role's complete data family remain out of
+scope. External sources become SourceSnapshots; reusable structured rows
+become immutable Datasets. Handoffs carry compact Snapshot/Dataset/Artifact IDs
+rather than raw source dumps. This is skill guidance, not a provider router or
+trust engine.
 
 ## Hook Boundary
 

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/docs/roles-skills-and-workflows.md

@@ -68,9 +68,10 @@ external package safety limits remain separate from this authoring rule.
 TradingCodex uses one lexical contract for managed, selection, and order
 skills. It removes one UTF-8 BOM, normalizes CRLF, CR, NEL, and Unicode
 line/paragraph separators to LF, tolerates leading and trailing blank lines,
-and trims horizontal whitespace for recognizing an invocation. It does not
-case-fold skill ids, normalize confusable characters, or ignore zero-width
-characters.
+and trims horizontal whitespace for recognizing an invocation. ASCII letter
+case in a skill id is presentation-only and resolves to the canonical lowercase
+id, which accommodates Codex app skill-chip rendering. The parser does not
+normalize Unicode lookalikes, confusable separators, or zero-width characters.
 
 A plain `$skill-id` and a Markdown link are equivalent only when the link label
 is that exact id and its target resolves to the matching projected
@@ -81,12 +82,14 @@ distinct managed markers remain invalid. Brain and Strategy selection accepts a
 projected link, deduplicates repeated references to the same id, and rejects
 distinct multiple ids.
 
-Order syntax is intentionally narrower. `$tcx-order-allow --mode
-paper|validation|live` occupies the complete first meaningful line and requires
-a non-empty request below it. Immediate submit/cancel remains an action-only
-prompt with exact `--name value` pairs. A matching projected Markdown skill
-link may replace only the skill token; it does not relax flags, values, or the
-single-effect rule. Grant, mandate, and audit bindings continue to hash the
+Order syntax is intentionally narrower than research selection, but whitespace
+layout is not an authority boundary. `$tcx-order-allow`, its `--mode
+paper|validation|live` argument, and the non-empty request may share a line or
+wrap across lines. Immediate submit/cancel remains an action-only prompt with
+literal `--name value` pairs, which may wrap across lines. A matching projected
+Markdown skill link may replace only the skill token; case and whitespace
+normalization do not relax flags, identifiers, duplicate detection, mode, or
+the single-effect rule. Grant, mandate, and audit bindings continue to hash the
 original unnormalized prompt.
 
 ## Fixed Team
@@ -210,18 +213,23 @@ only current-workflow artifact candidates and Dataset cards/manifests. It
 assigns one producing owner and gives relevant children briefs that name that
 owner, exact reusable IDs, and the needed or missing slice; roles do not browse
 a whole catalog. A stale, incomplete, or mismatched record still contributes
-its valid portion, while only the owner retrieves missing coverage. Head Manager
-then gives the evidence producer the smallest missing-data question and other
-roles compact SourceSnapshot, Dataset, and ResearchArtifact IDs. The producer
-follows the canonical `tcx-source-gate` procedure: reuse supplied evidence, use
-one relevant enabled user capability, use optional direct OpenBB, prefer
-original public records, then reliable web sources, and state a remaining gap.
+its valid portion. The producing owner may autonomously collect additional
+evidence when a newly discovered gap or conflict could materially change its
+assigned answer, readiness, or confidence and remains within the user's scope
+and that role's specialty; it does not need a Head Manager follow-up naming
+every field or source. Other roles receive compact SourceSnapshot, Dataset, and
+ResearchArtifact IDs. The producer follows the canonical `tcx-source-gate`
+preference: reuse supplied evidence, use one relevant enabled user capability,
+use optional direct OpenBB, prefer original public records, then reliable web
+sources, and state a remaining gap.
 
 This is agent guidance, not a Django routing state machine. The role names the
 provider where possible, does not repeat unchanged calls to the same source,
-and preserves a partial result while seeking only the missing field, identifier,
-or period. User capabilities remain BYOR; TradingCodex does not install,
-classify, proxy, approve, or audit them.
+and preserves a partial result while seeking unsupported coverage rather than
+refetching adequate coverage. There is no fixed source, search, or tool-call
+count. Broad just-in-case collection and recreation of another role's complete
+data family remain out of scope. User capabilities remain BYOR; TradingCodex
+does not install, classify, proxy, approve, or audit them.
 
 Use `record_source_snapshot` for every external source. Create a Dataset only
 for reusable structured rows, and bind the resulting Snapshot and Dataset IDs to
@@ -547,8 +555,8 @@ consumer; without proof injected by `PreToolUse` for the current exact
 `$tcx-order-allow` turn, it has no execution authority.
 
 For already-known canonical identifiers, the final external effect remains
-available from a root native Codex user turn whose meaningful content matches
-one exact action-only skill invocation. The canonical plain forms are:
+available from a root native Codex user turn whose meaningful content contains
+one action-only skill invocation. The canonical single-line forms are:
 
 ```text
 $tcx-order-submit --ticket-id <ticket-id> --approval-receipt-id <approval-receipt-id>
@@ -557,8 +565,9 @@ $tcx-order-cancel --ticket-id <ticket-id> --broker-order-id <broker-order-id> --
 $tcx-order-cancel --ticket-id <ticket-id> --broker-order-id <broker-order-id> --approval-receipt-id <approval-receipt-id> --live-confirmation <token>
 ```
 
-The leading skill token may be its matching projected Markdown link; every
-other part of the grammar remains literal. These two root skill bundles
+The leading skill token is ASCII case-insensitive and may be its matching
+projected Markdown link. Whitespace, including normalized line breaks, may
+separate tokens; every other part of the grammar remains literal. These two root skill bundles
 describe a protocol and disable implicit invocation; they carry no MCP or
 broker authority. `UserPromptSubmit` recognizes the reserved invocation,
 rejects malformed or subagent forms,
@@ -567,7 +576,7 @@ parses the complete prompt deterministically, creates a workspace-bound
 an analysis run begins.
 
 For a workflow that creates or selects identifiers during the turn, the first
-meaningful line must instead contain exactly one canonical mode invocation:
+meaningful invocation must instead be the order-allow skill:
 
 ```text
 $tcx-order-allow --mode paper
@@ -575,9 +584,10 @@ $tcx-order-allow --mode validation
 $tcx-order-allow --mode live
 ```
 
-The skill token may be its matching projected Markdown link, but no prose or
-extra flag may share the line and a non-empty workflow request must follow. The
-hook requires a root native Codex `session_id` and `turn_id`, issues one grant
+The skill token may be its matching projected Markdown link. Its mode and
+non-empty workflow request may follow on the same line or later lines; line
+layout does not change the grant. The hook requires a root native Codex
+`session_id` and `turn_id`, issues one grant
 bound to workspace, session, turn, original complete prompt hash, Codex permission
 mode, and execution mode, then continues ordinary orchestration. Plan mode
 rejects immediate order effects plus grant issuance and use. The grant expires

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/docs/safety-policy-and-execution.md

@@ -18,22 +18,23 @@ raise quality without becoming executable authorization.
 Every final submit or cancel action follows one of two native-user entries:
 
 ```text
-exact immediate root action -> deterministic hook parse -> native-user permission
+literal action-only root invocation -> deterministic hook parse -> native-user permission
   -> policy -> payload validation -> approval/duplicate-request check
   -> mandatory intent audit -> connection -> finalized/uncertain audit
 
-exact first-meaningful-line $tcx-order-allow -> workspace/session/turn/prompt/mode grant
+first-meaningful $tcx-order-allow invocation -> workspace/session/turn/prompt/mode grant
   -> canonical workflow state -> protected tool + PreToolUse proof
   -> consume once -> the same policy/approval/idempotency/live/audit gates
 ```
 
 This order matters:
 
-1. Exact user admission: require either one action-only immediate submit/cancel
-   prompt or an exact first meaningful line `$tcx-order-allow --mode
-   paper|validation|live` on the current root turn. A matching projected
-   Markdown skill link may replace only the skill token; all arguments stay
-   literal.
+1. Explicit user admission: require either one action-only immediate
+   submit/cancel prompt or a first-meaningful `$tcx-order-allow` invocation with
+   `--mode paper|validation|live` and a non-empty request on the current root
+   turn. A matching projected Markdown skill link may replace only the skill
+   token. ASCII skill-name case and whitespace layout are presentation details;
+   all arguments stay literal.
 2. Parse: create a workspace-bound immediate mandate, or a workspace-,
    session-, turn-, full-prompt-, and mode-bound single-use turn grant before
    any model or subagent runs.
@@ -69,13 +70,14 @@ no local-loopback mutation exception. Remote mutation use always requires
 staff/API-key authentication. Loopback is not generic mutation authority.
 
 Immediate final submit and cancel do not use an agent MCP or REST identity. The
-`UserPromptSubmit` hook accepts only an exact root native action prompt, binds a
+`UserPromptSubmit` hook accepts only a literal action-only root native prompt, binds a
 short-lived mandate to the original prompt hash and workspace provenance, and
 calls the canonical execution gateway in-process as `native-user`.
 
 For a workflow that creates or selects canonical identifiers later in the
-turn, the same hook accepts only an exact first-meaningful-line
-`$tcx-order-allow` mode, requires current Codex `session_id` and `turn_id`, and issues a grant
+turn, the same hook accepts a first-meaningful `$tcx-order-allow` invocation,
+mode, and non-empty request across arbitrary normalized whitespace, requires
+current Codex `session_id` and `turn_id`, and issues a grant
 bound to workspace, session, turn, original full prompt hash, Codex permission mode, and
 execution mode. Plan mode rejects both immediate execution and turn-grant
 issuance or use; the user must start a new non-Plan root turn. The grant expires
@@ -579,19 +581,20 @@ has an approval receipt, and `submit_approved_order` includes
   $tcx-order-cancel --ticket-id <ticket-id> --broker-order-id <broker-order-id> --approval-receipt-id <approval-receipt-id> --live-confirmation <token>
   ```
 
-- The deterministic parser accepts only literal `--name value` pairs. It rejects
+- The deterministic parser accepts only literal `--name value` pairs, while
+  allowing normalized whitespace and line breaks between tokens. It rejects
   prose, aliases, quotes, escapes, comments, duplicate or unknown flags,
   `--name=value`, multiple actions, subagent turns, and the
   retired `$execute-paper-order` invocation. A malformed prompt beginning with a
   reserved token is blocked rather than falling through to analysis. One UTF-8
-  BOM, normalized line endings, leading/trailing blank lines, and a Markdown
+  BOM, ASCII skill-name case, normalized line endings, leading/trailing blank lines, and a Markdown
   link whose label and target match the projected action skill do not relax
   that action-only grammar.
 - The two root skill bundles document this protocol but carry no tool authority.
   The hook creates the in-memory mandate from the original user prompt, records
   a redacted prompt hash and normalized identifiers, and calls the Django
   gateway synchronously before normal Head Manager orchestration.
-- A workflow may instead put one exact invocation on its first meaningful line:
+- A workflow may instead begin with one order-allow invocation:
 
   ```text
   $tcx-order-allow --mode paper
@@ -599,8 +602,9 @@ has an approval receipt, and `submit_approved_order` includes
   $tcx-order-allow --mode live
   ```
 
-  The skill token may be its matching projected Markdown link. The remainder
-  must contain a non-empty normal interactive or Scheduled Task prompt. The line
+  The skill token may be its matching projected Markdown link. The mode and
+  non-empty normal interactive or Scheduled Task request may share the first
+  meaningful line or wrap across later lines. This invocation
   admits at most one later submit or cancel and does not itself perform an
   effect. A free-form mention, later-line token, malformed mode, or subagent
   turn grants nothing. The browser viewer exposes no prompt entrypoint. Scheduled and interactive prompts use the
@@ -621,8 +625,9 @@ has an approval receipt, and `submit_approved_order` includes
   materially ambiguous, Head Manager asks the user or stops at `waiting` or
   `blocked`.
 - Order drafting and approval retain their role-owned structured tools. Final
-  cancellation and execution additionally require either the exact immediate
-  native grammar and parser-issued mandate or the exact first-meaningful-line turn grant
+  cancellation and execution additionally require either the literal
+  action-only native grammar and parser-issued mandate or the first-meaningful
+  turn grant
   plus protected proof, together with every deterministic principal,
   capability, policy, approval, idempotency, connection, and audit gate.
   Analytical prose cannot activate those paths.

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_build_turn_grant.py

@@ -107,7 +107,7 @@ def test_build_parser_accepts_first_meaningful_line_and_workspace_links(workspac
     for prompt in accepted:
         assert parse_build_invocation(prompt, workspace) is True
     assert parse_build_invocation("Research how $tcx-build works without changing files.") is None
-    assert parse_build_invocation("$TCX-build\nUpdate the provider.") is None
+    assert parse_build_invocation("$TCX-build\nUpdate the provider.") is True
     assert parse_build_invocation("\u200b$tcx-build\nUpdate the provider.") is None
     assert parse_build_invocation("$tcx-buіld\nUpdate the provider.") is None
 

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_codex_native_orchestration.py

@@ -70,7 +70,10 @@ def test_fixed_role_prompts_use_natural_evidence_distinctions() -> None:
     assert "Keep disposable work under `$TRADINGCODEX_SCRATCH`" in flat_base
     assert "never read audit records" in flat_base
     assert "treat the new request as a bounded delta" in flat_base
-    assert "retrieve only missing coverage" in flat_base
+    assert "collect additional evidence whenever a newly discovered gap or conflict" in flat_base
+    assert "do not need a follow-up from Head Manager naming every field or source first" in flat_base
+    assert "there is no fixed search or tool-call count" in flat_base
+    assert "retrieve the coverage needed to resolve the delta" in flat_base
     assert "answers the exact assigned question" in flat_base
     assert "If the requested delta belongs to another specialty" in flat_base
     for prompt in [base, *role_prompts, *role_skills]:
@@ -166,7 +169,7 @@ def test_child_briefs_and_artifact_skills_keep_capabilities_and_lineage_distinct
     assert "triggering cross-role Artifact IDs (inputs)" in flat_fixed
     assert "create a new artifact only when the brief explicitly says so" in flat_fixed
     assert "Honor the brief's data-family owner" in flat_fixed
-    assert "do not recollect another role's complete family" in flat_fixed
+    assert "Do not recollect another role's complete family" in flat_fixed
     assert "Do not load `tcx-calculation` merely to quote or compare source-reported figures" in flat_fixed
 
     assert "brief every child with that ownership, exact reusable IDs, and the missing slice" in flat_workflow

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_investment_brain_provenance.py

@@ -123,6 +123,12 @@ def _research_args(artifact_id: str, artifact_type: str, run_id: str, *, inputs:
 
 def test_explicit_brain_invocation_is_exact_and_single() -> None:
     assert explicit_investment_brain_invocation(f"Use ${BRAIN_ID}.") == BRAIN_ID
+    assert explicit_investment_brain_invocation(
+        "$Investment-Brain-Quality-Growth\nAnalyze ACME."
+    ) == BRAIN_ID
+    assert explicit_investment_brain_invocation(
+        f"${BRAIN_ID}\n$strategy-quality-watch\nAnalyze ACME."
+    ) == BRAIN_ID
     assert explicit_investment_brain_invocation(f"Use {BRAIN_ID}.") == ""
     assert explicit_investment_brain_invocation(f"${BRAIN_ID} then ${BRAIN_ID}") == BRAIN_ID
     with pytest.raises(ValueError, match="exactly one"):
@@ -159,6 +165,10 @@ def test_explicit_brain_invocation_accepts_only_the_projected_skill_link(
     prompt = f"Use [${BRAIN_ID}]({target}) and ${BRAIN_ID}."
 
     assert explicit_investment_brain_invocation(prompt, workspace) == BRAIN_ID
+    assert explicit_investment_brain_invocation(
+        f"Use [$Investment-Brain-Quality-Growth]({target}).",
+        workspace,
+    ) == BRAIN_ID
     with pytest.raises(ValueError, match="must target"):
         explicit_investment_brain_invocation(
             f"Use [${BRAIN_ID}]({workspace / '.agents/skills/tcx-build/SKILL.md'}).",

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_investor_context.py

@@ -132,6 +132,12 @@ def test_analysis_run_seals_enabled_context_and_respects_workspace_default(tmp_p
 
 def test_explicit_strategy_invocation_never_guesses_from_natural_language() -> None:
     assert explicit_strategy_invocation("Use $strategy-quality-watch for this review.") == "strategy-quality-watch"
+    assert explicit_strategy_invocation(
+        "$Strategy-Quality-Watch\nReview this portfolio."
+    ) == "strategy-quality-watch"
+    assert explicit_strategy_invocation(
+        "$investment-brain-quality-growth\n$strategy-quality-watch\nReview this portfolio."
+    ) == "strategy-quality-watch"
     assert explicit_strategy_invocation("Use strategy-quality-watch for this review.") == ""
     assert explicit_strategy_invocation("Use $tcx-strategy to create a reusable strategy.") == ""
     with pytest.raises(ValueError, match="exactly one"):
@@ -146,6 +152,10 @@ def test_explicit_strategy_invocation_accepts_only_the_projected_skill_link(tmp_
 
     assert explicit_strategy_invocation(prompt, tmp_path) == "strategy-quality-watch"
     assert explicit_strategy_invocation(f"{prompt} $strategy-quality-watch", tmp_path) == "strategy-quality-watch"
+    assert explicit_strategy_invocation(
+        f"[$Strategy-Quality-Watch]({skill})\nReview it.",
+        tmp_path,
+    ) == "strategy-quality-watch"
     with pytest.raises(ValueError, match="must target"):
         explicit_strategy_invocation(
             f"Use [$strategy-quality-watch]({tmp_path / 'other/SKILL.md'}).",

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_native_execution_gateway.py

@@ -145,6 +145,26 @@ def test_native_invocation_accepts_formatting_and_exact_workspace_link(workspace
     assert mandate.prompt_sha256 == hashlib.sha256(prompt.encode()).hexdigest()
 
 
+@pytest.mark.parametrize(
+    "prompt",
+    [
+        "$TCX-ORDER-SUBMIT --ticket-id ticket-1 --approval-receipt-id receipt-1",
+        "$tcx-order-submit\n--ticket-id ticket-1\n--approval-receipt-id receipt-1",
+        "$tcx-order-submit --ticket-id ticket-1\n--approval-receipt-id receipt-1",
+    ],
+)
+def test_native_invocation_accepts_case_and_line_wrapping(
+    workspace: Path,
+    prompt: str,
+) -> None:
+    mandate = parse_native_execution_invocation(prompt, workspace)
+
+    assert mandate is not None
+    assert mandate.action == NATIVE_SUBMIT_ACTION
+    assert mandate.ticket_id == "ticket-1"
+    assert mandate.approval_receipt_id == "receipt-1"
+
+
 def test_native_invocation_rejects_mismatched_workspace_link(workspace: Path) -> None:
     prompt = (
         f"[$tcx-order-submit]({workspace / '.agents/skills/tcx-order-cancel/SKILL.md'}) "
@@ -166,7 +186,6 @@ def test_native_invocation_rejects_mismatched_workspace_link(workspace: Path) ->
         "$tcx-order-submit ticket-1 --approval-receipt-id receipt-1",
         "$tcx-order-submit --ticket-id 'ticket-1' --approval-receipt-id receipt-1",
         "$tcx-order-submit --ticket-id ticket\\-1 --approval-receipt-id receipt-1",
-        "$tcx-order-submit --ticket-id ticket-1\n--approval-receipt-id receipt-1",
         "$tcx-order-submit --ticket-id ticket-1 --approval-receipt-id receipt-1 extra",
         "$tcx-order-submit --ticket-id ticket-1 --approval-receipt-id receipt-1 "
         "$tcx-order-cancel --ticket-id ticket-1",

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_order_turn_grant.py

@@ -144,6 +144,9 @@ def test_order_allow_parser_accepts_format_only_variants(workspace: Path, mode:
     target = workspace / ".agents/skills/tcx-order-allow/SKILL.md"
     prompts = (
         order_allow_prompt(mode),
+        f"$TCX-ORDER-ALLOW --mode {mode}\nRun the requested task.",
+        f"$tcx-order-allow --mode {mode} Run the requested task.",
+        f"$tcx-order-allow\n--mode\n{mode}\nRun the requested task.",
         f"$tcx-order-allow --mode {mode}\r\nRun the requested scheduled task.",
         f"\ufeff\n \t$tcx-order-allow --mode {mode} \t\u2028Run the requested task.\n",
         f"\u00a0$tcx-order-allow\u3000--mode {mode}\u00a0\nRun the requested task.",
@@ -161,7 +164,6 @@ def test_order_allow_parser_accepts_format_only_variants(workspace: Path, mode:
         "$tcx-order-allow --mode paper\n\ufeff",
         "$tcx-order-allow",
         "$tcx-order-allow --mode=paper\nSubmit it.",
-        "$tcx-order-allow --mode PAPER\nSubmit it.",
         "$tcx-order-allow --mode paper --extra yes\nSubmit it.",
         "$tcx-order-allow --mode paper\n$tcx-build\nChange the connector.",
         "$tcx-order-allow --mode paper\n\u200b$tcx-build\nChange the connector.",

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_release_contract.py

@@ -118,7 +118,7 @@ def test_cli_hook_dispatch_preserves_standard_input_and_output(
 def test_v2_package_metadata_has_one_stable_version_source() -> None:
     project = tomllib.loads((ROOT / "pyproject.toml").read_text(encoding="utf-8"))
 
-    assert TRADINGCODEX_VERSION == "2.1.1"
+    assert TRADINGCODEX_VERSION == "2.1.2"
     assert str(Version(TRADINGCODEX_VERSION)) == TRADINGCODEX_VERSION
     assert project["project"]["dynamic"] == ["version"]
     assert "version" not in project["project"]

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tests/test_source_snapshot_agent_contract.py

@@ -203,6 +203,10 @@ def test_source_snapshot_api_tool_and_role_instructions_align() -> None:
     assert "Treat\nOpenBB as access to its returned provider" in source_gate
     assert "Secondary does not mean screen-only" in source_gate
     assert "do not require a fixed source count" in source_gate
+    assert "not a one-call budget" in source_gate
+    assert "newly discovered gap, contradiction, stale anchor, or\nidentifier problem" in source_gate
+    assert "number and mix of useful sources" in source_gate
+    assert "another source is appropriate when\nit resolves a material gap or conflict" in source_gate
     assert "Use `evidence_readiness: decision-grade` when every\nconclusion-driving claim" in source_gate
     assert "not merely because a primary source is\nabsent" in source_gate
     assert "latest completed period can be the current usable anchor" in source_gate
@@ -222,6 +226,8 @@ def test_source_snapshot_api_tool_and_role_instructions_align() -> None:
     assert "SourceSnapshot" in openbb_reference and "Dataset" in openbb_reference
     for field in ("snapshot_id", "retrieved_at", "recorded_at", "known_at"):
         assert f"`{field}`" in collect_evidence
+    assert "without waiting for a new field-by-field\n   instruction" in collect_evidence
+    assert "Use evidence value, not a fixed source or call count" in collect_evidence
 
     artifact_cutoff = TOOL_REGISTRY["create_research_artifact"].input_schema[
         "properties"

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tradingcodex.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tradingcodex
-Version: 2.1.1
+Version: 2.1.2
 Summary: Local-first investment operating system for Codex-native research and service-gated execution.
 Author: TradingCodex Authors
 License-Expression: Apache-2.0

{tradingcodex-2.1.1 → tradingcodex-2.1.2}/tradingcodex_service/application/execution_gateway.py

@@ -46,7 +46,10 @@ NATIVE_TURN_GRANT_SOURCE = "native_codex_turn_grant"
 ORDER_TURN_GRANT_TTL = timedelta(hours=1)
 ORDER_TURN_GRANT_MODES = frozenset({"paper", "validation", "live"})
 ORDER_TURN_GRANT_ACTIONS = frozenset({"submit", "cancel"})
-_ORDER_ALLOW_LINE = re.compile(r"^\$tcx-order-allow --mode (paper|validation|live)$")
+_ORDER_ALLOW_BODY = re.compile(
+    r"^--mode[ \t\n]+(?P<mode>paper|validation|live)[ \t\n]+(?P<request>.+)$",
+    re.ASCII | re.DOTALL | re.IGNORECASE,
+)
 
 _RESERVED_SKILL_TOKENS = frozenset(
     {
@@ -192,9 +195,12 @@ def parse_native_execution_invocation(
     if len(raw.encode("utf-8")) > _MAX_PROMPT_BYTES:
         raise NativeExecutionInvocationError("native execution invocation is too long")
     lines = meaningful_lines(raw)
-    if len(lines) != 1:
-        raise NativeExecutionInvocationError("native execution invocation must be one action-only line")
-    value = token + (f" {invocation.tail}" if invocation.tail else "")
+    remaining_lines = [line for _, line in lines[1:]]
+    value = " ".join(
+        part
+        for part in (token, invocation.tail, *remaining_lines)
+        if part
+    )
     if any(_is_control_character(character) for character in value):
         raise NativeExecutionInvocationError("native execution invocation contains a control character")
     if any(character in value for character in ("'", '"', "\\")):
@@ -253,7 +259,7 @@ def parse_order_allow_invocation(
     prompt: Any,
     workspace_root: Path | str | None = None,
 ) -> str | None:
-    """Return a first-meaningful-line turn-grant mode for an exact invocation."""
+    """Return a turn-grant mode from a reserved first invocation."""
 
     raw = raw_prompt(prompt)
     try:
@@ -266,14 +272,22 @@ def parse_order_allow_invocation(
         raise NativeExecutionInvocationError(str(exc)) from exc
     if invocation is None:
         return None
-    match = _ORDER_ALLOW_LINE.fullmatch(f"{ORDER_ALLOW_SKILL} {invocation.tail}")
+    lines = meaningful_lines(raw)
+    body = "\n".join(
+        part
+        for part in (invocation.tail, *(line for _, line in lines[1:]))
+        if part
+    )
+    match = _ORDER_ALLOW_BODY.fullmatch(body)
     if match is None:
         raise NativeExecutionInvocationError(
-            "$tcx-order-allow must use exact syntax: $tcx-order-allow --mode paper|validation|live"
+            "$tcx-order-allow requires --mode paper|validation|live followed by a non-empty request"
         )
-    lines = meaningful_lines(raw)
-    if len(lines) < 2 or not any(has_visible_content(line) for _, line in lines[1:]):
-        raise NativeExecutionInvocationError("$tcx-order-allow requires a non-empty request after its invocation")
+    request = match.group("request")
+    if not has_visible_content(request):
+        raise NativeExecutionInvocationError("$tcx-order-allow requires a non-empty request")
+    if request.lstrip().startswith("--"):
+        raise NativeExecutionInvocationError("$tcx-order-allow contains an unsupported option")
     incompatible = (
         "$tcx-build",
         "$tcx-brain",
@@ -287,7 +301,7 @@ def parse_order_allow_invocation(
         mixed = next(
             (
                 found
-                for _, line in lines[1:]
+                for _, line in meaningful_lines(request)
                 if (
                     found := parse_line_invocation(
                         line,
@@ -306,7 +320,7 @@ def parse_order_allow_invocation(
         )
     if len(raw.encode("utf-8")) > _MAX_TURN_GRANT_PROMPT_BYTES:
         raise NativeExecutionInvocationError("$tcx-order-allow prompt is too long")
-    return match.group(1)
+    return match.group("mode").lower()
 
 
 def issue_order_turn_grant(