aeat.agent package¶
Operator agent-harness: read accessor for the shipped operating layer.
The agent-harness operating layer - operator rules, tax-advisor personas, and
workflow skills - is reviewed markdown product data under aeat/_data/agent/,
shipped inside the wheel and read here through the bundled-data boundary so it
resolves identically under an editable install and a built wheel.
This package is a read accessor only. It carries no tax logic and computes no
value; it hands the operating-layer text to whatever drives the agent (a prompt
assembler, the MCP server, or an operator-workspace materialiser). The capability
catalogue the agent reads first is emitted by aeat app contract --format json.
- iter_operator_rules()[source]¶
Yield each operator operating-rule document, ordered by file name.
- Return type:
- operator_rules_text()[source]¶
Return the concatenated operator operating-rule documents.
The rules are joined in file-name order with a blank line between them, ready to load into an agent’s always-on operating context.
- Return type:
- iter_personas()[source]¶
Yield each tax-advisor persona document, ordered by file name.
- Return type:
- iter_skill_documents()[source]¶
Yield each workflow skill’s
SKILL.mddocument, ordered by skill name.- Return type:
- iter_skill_metadata()[source]¶
Yield every shipped skill’s parsed
SKILL.mdmetadata.Each frontmatter is parsed through the structured schema. A predicate that is present is fully validated: this raises
SkillMetadataErroron the first skill whose frontmatter is malformed, whoseapplies_whenpredicate is invalid, or whose declared name does not match its directory. A skill whose predicate has not yet been lifted from prose still loads, withapplies_whenset toNone- strict presence is enforced by the coverage gate, not this load path, so the tree stays loadable while the lifts land.- Return type:
- Returns:
Subpackages¶
- aeat.agent.eval package
- Submodules
- aeat.agent.eval._flywheel module
- aeat.agent.eval._live_harness module
- aeat.agent.eval._live_scoring module
- aeat.agent.eval._models module
GoldenScenarioNarrationFaithfulnessConfirmationTierConfirmationGateCheckGoldenResultExitCodeScenarioExitCodeVerdictUnderDeclarationScenarioUnderDeclarationVerdictContradictionScenarioContradictionVerdictProfileConfirmationScenarioProfileConfirmationVerdictElicitationActionLiveToolCallRecordLiveNarrationRecordLiveElicitationRecordLiveTrajectoryLiveInvariantVerdict
- aeat.agent.eval._replay module
- aeat.agent.eval._report module
- aeat.agent.eval._runner module
- Submodules
Submodules¶
- aeat.agent._skill_metadata module
- aeat.agent._workspace module