awslabs/iam-policy-autopilot 0.4.0 on GitHub

Policy split size is configurable through --max-policy-size, the max_policy_size library setting, the MaxPolicySize MCP input, and the maxPolicySize WASM option. It accepts non-negative values and defaults to 6,144 non-whitespace characters. Statements that do not fit start a new policy; oversized single-statement policies are emitted with a warning.
The opt-in --allow-overpermissive-wildcards-to-limit-total-output-size target can introduce word-aligned action wildcards when exact policies exceed the requested combined size. Equivalent library, MCP, WASM, TypeScript, and telemetry fields are included. Wildcard statements and unmet targets include structured warnings.

Added file path safety guidance to the MCP server instructions, the generate_application_policies tool description, and the Kiro power: agents must resolve input paths to canonical absolute paths, verify they are within the user's workspace, and reject paths containing shell metacharacters. Agent runtimes and hosting environments that invoke the CLI directly must use array-based subprocess execution instead of shell string interpolation.

Policy statements with identical effects, actions, resources, and a shared condition operator/key now merge their condition values into one sorted, deduplicated list.
Java SDK v2 calls whose operation names contain consecutive capitals, such as rds.createDBCluster() and rds.describeDBClusters(), are now included in generated policies.
AWS profiles configured with aws login now work with credential-dependent commands such as --upload-policies and fix-access-denied. (#276)