Llm Extension¶
LLM Extension Workflows (spec: llm_bootstrap_engineering.md).
Cluster: Uncategorised | Type: component | MCP Tools: None
Overview¶
Public API¶
LlmExtensionContract¶
Read-only facade over the existing deterministic rollout gate.
Methods:
assess(manifest: ExtensionManifest, evals: EvalReport) -> RolloutDecision¶
ExtensionManifest¶
§6.3 extension manifest — the rollout-decision-relevant fields.
| Field | Type | Default |
|---|---|---|
extension_id | str | '' |
requested_rollout_mode | str | SHADOW |
required_pass_rate | float | 0.92 |
security_boundary | bool | False |
injection_scan_enabled | bool | False |
least_privilege | bool | False |
high_impact_tools_approval_gated | bool | True |
rollback_path | str | '' |
version | str | '' |
observability | bool | False |
EvalReport¶
The result of running the extension's eval suite (§20.2). Produced by the existing
| Field | Type | Default |
|---|---|---|
pass_rate | float | 0.0 |
required_types_present | bool | False |
failed_eval_types | list[str] | field(default_factory=list) |
constraint_complete | bool | False |
regression_tests_present | bool | False |
RolloutDecision¶
The staged-rollout verdict (the family's centerpiece).
| Field | Type | Default |
|---|---|---|
granted_mode | str | required |
requested_mode | str | required |
max_earned_mode | str | required |
reliability_label | str | required |
requires_human | bool | required |
capped | bool | required |
blockers | list[str] | field(default_factory=list) |
reasons | list[str] | field(default_factory=list) |
Methods:
to_dict() -> dict¶
Functions¶
assess_rollout_readiness(manifest: ExtensionManifest, evals: EvalReport) -> RolloutDecision¶
Grant the staged-rollout mode EARNED by the evidence. DOWN-only: granted =