tidaldb/.sdlc/features/m10-agent-capability-boundaries/qa-plan.md

QA Plan: Agent Capability Boundaries

Scope

This QA plan covers the policy enforcement layer added by m10-agent-capability-boundaries. All testing is automated via the Rust test suite. No manual or UI testing is required — this feature has no frontend surface.

Test Levels

1. Unit Tests (cargo test --lib)

All unit tests live in #[cfg(test)] blocks within the modules they test.

session/policy.rs — PolicyEvaluator read checks

Test	Description	Expected
check_read_empty_lists_allows_any	Empty allowed_read_signals and denied_read_signals	Ok(()) for any signal
check_read_allow_list_permits_listed_signal	allowed_read_signals: ["view"], check "view"	Ok(())
check_read_allow_list_rejects_unlisted_signal	allowed_read_signals: ["view"], check "like"	Err(ReadNotAllowed)
check_read_deny_list_rejects_denied_signal	denied_read_signals: ["hide"], check "hide"	Err(ReadDenied)
check_read_deny_list_takes_priority_over_allow	allowed: ["hide"], denied: ["hide"], check "hide"	Err(ReadDenied)
check_attribute_read_empty_lists_allows_any	All attribute lists empty	Ok(()) for any key
check_attribute_read_allow_list_permits_key	allowed_user_attributes: ["locale"], key "locale"	Ok(())
check_attribute_read_allow_list_rejects_unlisted_key	allowed_user_attributes: ["locale"], key "age_range"	Err(AttributeReadNotAllowed)
check_attribute_read_deny_list_rejects_key	denied_user_attributes: ["email"], key "email"	Err(AttributeReadDenied)
check_profile_override_empty_list_rejects_any	allowed_profile_overrides: [], profile "search"	Err(ProfileOverrideNotAllowed)
check_profile_override_listed_profile_allowed	allowed_profile_overrides: ["search"], profile "search"	Ok(())
check_profile_override_unlisted_profile_rejected	allowed_profile_overrides: ["search"], profile "for_you"	Err(ProfileOverrideNotAllowed)

schema/validation/builders.rs — Schema build validation

Test	Description	Expected
schema_policy_valid_read_signals_builds	Valid signal names in read lists	Ok(schema)
schema_policy_unknown_signal_in_allowed_read_fails	allowed_read_signals: ["nonexistent"]	Err(SchemaError)
schema_policy_unknown_signal_in_denied_read_fails	denied_read_signals: ["nonexistent"]	Err(SchemaError)
schema_policy_conflict_allow_deny_read_fails	Same signal in both lists	Err(SchemaError)
schema_policy_unknown_profile_override_fails	allowed_profile_overrides: ["made_up_profile"]	Err(SchemaError)
schema_policy_star_sentinel_expands_to_all_profiles	allowed_profile_overrides: ["*"]	Schema builds; list contains all known profiles
schema_policy_empty_new_fields_is_valid	All new fields empty	Ok(schema) (no regression)

session/state.rs + session/snapshot.rs

Test	Description	Expected
overrides_rejected_starts_at_zero	Freshly created SessionState	overrides_rejected.load() == 0
snapshot_includes_overrides_rejected	Snapshot from state with overrides_rejected = 3	snapshot.overrides_rejected == 3
frozen_snapshot_includes_overrides_rejected	Frozen snapshot from same state	frozen_snapshot.overrides_rejected == 3

2. Integration Tests (tidal/tests/m10_agent_capability.rs)

These tests exercise the full TidalDb API stack. All run with an ephemeral in-memory database.

Test	Description	Expected
read_allowed_signal_succeeds	Session with allowed_read_signals: ["view"]; call read_decay_score_for_session for view	Ok(Some(_))
read_disallowed_signal_fails_not_allowed	Same session; call for like (not in allow list)	Err with ReadNotAllowed kind
read_denied_signal_fails_denied	Session with denied_read_signals: ["hide"]; call for hide	Err with ReadDenied kind
read_without_session_unrestricted	read_decay_score (no session) for hide	Ok(_)
attribute_read_allowed_succeeds	Session with allowed_user_attributes: ["locale"]; read locale	Ok(Some(_))
attribute_read_disallowed_fails	Same session; read age_range	Err with AttributeReadNotAllowed
profile_override_allowed_proceeds	Session with allowed_profile_overrides: ["search"]; retrieve with profile: "search"	Ok(results)
profile_override_disallowed_fails	Same session; retrieve with profile: "for_you"	Err with ProfileOverrideNotAllowed
overrides_rejected_incremented	After one disallowed profile override	session_snapshot(sid).overrides_rejected == 1
audit_log_records_read_denial	After a ReadDenied violation	session_audit(sid) contains entry with appropriate kind
signals_rejected_incremented_on_read_denial	After a read denial	session_snapshot(sid).signals_rejected >= 1
empty_policy_no_regression	Policy with all new fields empty; reads of any signal	Ok(_) (same as before)
schema_build_fails_unknown_read_signal	Policy with allowed_read_signals: ["ghost"]	SchemaError at build time
schema_build_fails_conflict	Same signal in both read allow and deny lists	SchemaError at build time
schema_sentinel_star_allows_all_profiles	allowed_profile_overrides: ["*"]; retrieve with profile: "for_you"	Ok(results)

3. Regression: Existing Test Suite

Run the full library test suite to verify no regressions:

cargo test --manifest-path tidal/Cargo.toml --lib

All tests passing before this feature must continue to pass. The new fields on AgentPolicy default to empty and must not change behavior for any existing code path.

Run existing integration test suites:

cargo test --manifest-path tidal/Cargo.toml --test m7_uat
cargo test --manifest-path tidal/Cargo.toml --test m5_uat
cargo test --manifest-path tidal/Cargo.toml --test m6_uat

These must all remain green.

Test Commands

# Full library unit tests (must all pass, no regressions)
cargo test --manifest-path tidal/Cargo.toml --lib

# New integration tests
cargo test --manifest-path tidal/Cargo.toml --test m10_agent_capability

# Lint and format
cargo fmt --manifest-path tidal/Cargo.toml --check
cargo clippy --manifest-path tidal/Cargo.toml -D warnings

# Prior UAT suites (regression check)
cargo test --manifest-path tidal/Cargo.toml --test m7_uat
cargo test --manifest-path tidal/Cargo.toml --test m5_uat
cargo test --manifest-path tidal/Cargo.toml --test m6_uat

Pass Criteria

• All unit tests in session/policy.rs, schema/validation/builders.rs, and session/snapshot.rs pass.
• All 15 integration tests in m10_agent_capability.rs pass.
• Full library test suite passes with no regressions.
• cargo clippy -D warnings is clean.
• cargo fmt --check is clean.
• Prior M5, M6, M7 UAT suites are all green.

Exclusions

• No manual browser or UI testing required (no UI surface).
• No load/benchmark testing required for this feature (enforcement cost is sub-100ns per check; no new hot paths).
• No WAL or storage format testing required (no on-disk changes).