Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@crates/uselesskey-bdd/Cargo.toml`:
- Line 19: The uk-all feature currently declares both
"uselesskey-bdd-steps/uk-all" and every individual sub-feature (e.g.,
"uselesskey-bdd-steps/uk-rsa", "uselesskey-bdd-steps/uk-ecdsa",
"uselesskey-bdd-steps/uk-ed25519", etc.), which is redundant; change the uk-all
feature definition to only include "uselesskey-bdd-steps/uk-all" so the umbrella
feature in the uselesskey-bdd-steps crate controls the sub-features (update the
uk-all array to contain only "uselesskey-bdd-steps/uk-all").

In `@crates/uselesskey-core-cache/Cargo.toml`:
- Line 18: The spin dependency in Cargo.toml includes an unused feature
"spin_mutex"; remove that feature so the dependency only enables the used
"mutex" feature. Edit the spin entry in Cargo.toml (the dependency declaration
for spin) to drop "spin_mutex" from the features array, leaving features =
["mutex"] to match the usage of spin::Mutex in lib.rs.

In `@crates/uselesskey-core-jwk/src/lib.rs`:
- Around line 23-28: The Display impl for Jwks currently calls
serde_json::to_string(...).expect(...), which can panic; change it to handle
serialization errors by mapping them into a fmt::Error and returning
Err(fmt::Error) (e.g., use serde_json::to_string(self).map_err(|_|
fmt::Error).and_then(|s| f.write_str(&s).map_err(|_| fmt::Error)) or match on
the Result and return Err(fmt::Error) on failure). Apply the same non-panicking
pattern to the other Display implementations mentioned (the other impl
fmt::Display blocks in this file) so none call expect() and all convert
serde_json errors to fmt::Error instead.
- Around line 23-28: The Display impls for Jwks, AnyJwk, and PrivateJwk
currently serialize full key material; add a clear rustdoc WARNING above each
Display implementation (impl fmt::Display for Jwks, impl fmt::Display for
AnyJwk, impl fmt::Display for PrivateJwk) stating that Display emits
secret/unredacted material and should not be used for logging or diagnostics,
and then provide a safe alternative by adding a public helper that returns a
redacted JSON (e.g., pub fn to_redacted_json(&self) -> String) which reuses the
redaction logic from the existing Debug impls; update the rustdoc to recommend
using Debug or to_redacted_json() for inspection.

In `@crates/uselesskey-core-negative/src/lib.rs`:
- Around line 271-292: The unbounded loops in find_pem_variant and
find_der_variant can hang; add an iteration cap (e.g., for i in 0..1_000) and
after the loop panic with a clear message indicating exhaustion (include target
and function name) so tests fail fast instead of looping forever; update both
functions (find_pem_variant and find_der_variant) to use the capped range and a
panic! if no matching variant is found.

In `@crates/uselesskey-core-seed/src/lib.rs`:
- Around line 28-37: Update the documentation for from_env_value to explicitly
state the 64-char edge case: when the trimmed input (after optional "0x" prefix)
is exactly 64 characters the function will attempt parse_hex_32 and return that
Err on invalid hex rather than falling back to blake3::hash; otherwise any other
string is hashed with blake3::hash. Reference from_env_value and parse_hex_32 in
the doc comment so callers know this behavior.

In `@crates/uselesskey-core-seed/tests/seed_prop.rs`:
- Around line 14-21: The test comment for
seed_from_env_value_non_64_lengths_are_ok is misleading: update the doc comment
to state the real invariant — that after trimming and stripping an optional "0x"
prefix (using trimmed and after_prefix) any input whose effective length
(after_prefix.len()) is not 64 should cause Seed::from_env_value(&s) to succeed;
mention that the strategy plus prop_assume! excludes 64-length effective inputs
and that non-hex characters in a 64-char string are not covered by this test.

In `@crates/uselesskey-core-x509-derive/Cargo.toml`:
- Around line 19-20: The rcgen = "0.13" and time = "0.3" dependency entries
should be removed from the uselesskey-core-x509-derive Cargo.toml and instead
added to the workspace Cargo.toml under [workspace.dependencies] so both crates
reuse the same pinned versions; locate the rcgen and time lines in the
uselesskey-core-x509-derive manifest (and the duplicate entries in
uselesskey-x509) and move them to the workspace dependencies section alongside
existing workspace = true entries like blake3 and rand_core, preserving the
exact version pins (rcgen = "0.13", time = "0.3") to avoid the rcgen 0.14
breaking changes.

In `@crates/uselesskey-core-x509-derive/src/lib.rs`:
- Around line 34-37: The doc comment for deterministic_base_time is off by one
day: with BASE_TIME_WINDOW_DAYS = 365 and using day_offset % 365 the computed
range is 2025-01-01 through 2025-12-31, not 2026-01-01; update the comment to
reflect the actual window or if you intended to include 2026-01-01 change
BASE_TIME_WINDOW_DAYS to 366 (and keep day_offset % BASE_TIME_WINDOW_DAYS) so
the range becomes 2025-01-01 through 2026-01-01; adjust the text near the
deterministic_base_time function accordingly and ensure references to
BASE_TIME_WINDOW_DAYS and the day_offset modulo logic match the chosen intent.

In `@crates/uselesskey-core-x509/src/negative.rs`:
- Around line 45-66: The X509Negative::SelfSignedButClaimsCA variant is
described as conflicting but the builder sets spec.is_ca = true and
spec.key_usage = KeyUsage::ca(), producing a consistent CA; change the fixture
to actually conflict or make naming/description accurate: either (A) modify the
SelfSignedButClaimsCA branch to introduce a conflict (e.g., keep spec.is_ca =
true but remove or clear key usage like not setting KeyUsage::ca(), or set an
explicit conflicting key usage), or (B) rename the variant/description in
description() to reflect that it produces a valid self-signed CA (update
X509Negative::SelfSignedButClaimsCA string) so behavior and description match;
update references accordingly (look for X509Negative::SelfSignedButClaimsCA,
spec.is_ca, KeyUsage::ca(), and the description() match arm).

In `@crates/uselesskey-jwk/Cargo.toml`:
- Around line 17-18: Replace the inline path+version dependency for
uselesskey-core-jwk with a workspace-managed dependency: add an entry for
uselesskey-core-jwk to the workspace's [workspace.dependencies] in the root
Cargo.toml (the same key and version that was used here) and then in the crate's
Cargo.toml (where the current line "uselesskey-core-jwk = { path = ..., version
= ... }" exists) change that dependency to use workspace = true (remove path and
version). This ensures the crate uses the central workspace version and avoids
silent divergence; update the dependency declaration in the crate that
references uselesskey-core-jwk to "uselesskey-core-jwk = { workspace = true }".

In `@crates/uselesskey-x509/Cargo.toml`:
- Around line 25-26: Remove the unused workspace dependencies by deleting the
rsa.workspace and base64.workspace entries from the crate's Cargo.toml;
specifically remove the lines declaring rsa.workspace = true and
base64.workspace = true so the crate no longer redundantly lists rsa (provided
transitively via uselesskey-rsa) and the unused base64 dependency.

In `@crates/uselesskey-x509/src/chain.rs`:
- Around line 311-317: Replace the silent cast of spec.rsa_bits to u32 with an
explicit try_from conversion to avoid silent truncation: call
u32::try_from(spec.rsa_bits).expect("rsa_bits fits u32") and then .to_be_bytes()
when building rsa_bits; update the rsa_bits binding used in
deterministic_base_time_from_parts and add the necessary TryFrom import if
required to resolve the trait.

In `@xtask/src/main.rs`:
- Around line 242-252: The PUBLISH_CRATES array (const PUBLISH_CRATES) is
missing the new crate "uselesskey-core-x509-spec"; update the PUBLISH_CRATES
constant to include "uselesskey-core-x509-spec" in the list so
publish-check/preflight and mutants include the new crate during release checks,
ensuring the string is added alongside the other "uselesskey-core-*" entries.

---

Outside diff comments:
In `@AGENTS.md`:
- Around line 77-90: The Workspace Structure list is missing several newly added
crate entries; update the Architecture → Workspace Structure list to include the
new crates by adding lines for uselesskey-core-seed, uselesskey-core-cache,
uselesskey-core-token, uselesskey-core-x509-derive, uselesskey-core-keypair,
uselesskey-core-kid, uselesskey-core-negative, uselesskey-core-sink,
uselesskey-feature-grid, and uselesskey-test-grid alongside the existing crate
entries (keep the same bullet style and ordering convention used in the list so
they are discoverable and consistent with entries like crates/uselesskey-core
and crates/uselesskey-x509).
- Line 1: The document header is incorrect: change the first-line heading from
"# CLAUDE.md" to a correct title for this file (for example "# AGENTS.md" or
another accurate heading that reflects the content) so the file AGENTS.md has
its own proper top-level heading; update only the heading line and keep the rest
of the content unchanged.

In `@crates/uselesskey-core-negative/src/lib.rs`:
- Around line 40-43: The match arm currently shadows the outer `bytes: &[u8;
32]` by binding a new `bytes: usize`; update the binding name in the `_` arm
(for example `truncate_chars` or `truncate_len`) and use that new name in the
call to `derived_truncate_len(pem, ...)` and when constructing
`CorruptPem::Truncate { bytes: ... }` so the slice `bytes` and the usize length
are not ambiguous; specifically edit the `_` arm around the calls to
`derived_truncate_len` and `corrupt_pem` to use the new identifier instead of
`bytes`.
- Line 17: The Truncate variant's field is misnamed: CorruptPem::Truncate {
bytes: usize } is interpreted as a character count by corrupt_pem and
corrupt_pem_deterministic (they use pem.chars().take(bytes)), so rename the
field to "chars" (CorruptPem::Truncate { chars: usize }) and update all uses in
corrupt_pem and corrupt_pem_deterministic to read the new field name;
alternatively, if you intended a byte count, change the logic to slice the PEM
by bytes using pem.as_bytes().take(bytes) and convert back safely—pick one
approach and make the variant, pattern matches, and consumers consistent.
- Around line 166-172: The truncation can return 0 and collapse for len==2;
update derived_truncate_len_bytes to mirror the PEM/derived_truncate_len
behavior by computing a value in [1, len-1] instead of [0, len-2]: keep the
early return for len <= 1, compute span = len - 1, and return 1 + (digest[2] as
usize % span) so at least one byte remains and len==2 yields 1; adjust any
comments if present in derived_truncate_len_bytes to reflect the new range.
- Around line 174-183: Add property-based tests using proptest to cover the
deterministic helpers' invariants: create a #[cfg(test)] proptest module and add
proptest cases that generate arbitrary byte slices and lengths to assert the
invariants for flip_byte (for any der and offset < der.len() assert exactly one
byte differs and new byte == der[offset] ^ 0x01), truncate_der (for any len <
der.len() assert output equals &der[..len]), derived_offset (for any len > 0 and
arbitrary selector assert result < len), and derived_truncate_len_bytes (for any
len > 1 assert result < len); reference the existing helper names flip_byte,
truncate_der, derived_offset, derived_truncate_len_bytes and use proptest
strategies for Vec<u8> and valid offsets/lengths to reproduce the conditions
described in the comment.

In `@crates/uselesskey-x509/src/chain.rs`:
- Around line 64-87: The new() constructor duplicates with_variant(...) by
hardcoding the "good" variant; refactor new to delegate to with_variant to
remove duplication: replace new's body so it calls Self::with_variant(factory,
label, spec, "good") (ensuring ownership/borrowing matches) rather than calling
load_chain_inner itself; update any signature/ownership adjustments as needed so
Factory, label (&str), and spec (ChainSpec) are passed correctly to
with_variant.

In `@xtask/src/plan.rs`:
- Around line 149-222: The dependents() graph is missing an entry for the
workspace crate "uselesskey-core-x509-spec", so changes there won't propagate to
dependents; update the match in the dependents() function to include a new arm
for "uselesskey-core-x509-spec" that lists its downstream crates (at minimum
"uselesskey-core-x509" and any other crates that directly depend on the spec),
ensuring the propagation includes this crate alongside the existing
"uselesskey-core-x509" entry.

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In `@crates/uselesskey-core-cache/Cargo.toml`:
- Line 18: The Cargo.toml currently enables an unused feature "spin_mutex" on
the spin dependency (spin = { version = "0.9.8", default-features = false,
features = ["mutex", "spin_mutex"] }); remove "spin_mutex" from the features
array, run the provided verification commands (searching for SpinMutex,
spin::SpinMutex, and spin::Mutex) to ensure no code references remain, then run
cargo build and cargo test (and cargo update if needed) to regenerate Cargo.lock
and confirm the change is safe.