refactor: split version checks into two typed release-assessment modules

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code

[Tryanks]

Architecture round, candidate C6.

Problem — version_check.rs mixed two release domains (provider CLIs, app GitHub releases) as helper fragments while providers.rs assembled both workflows itself; the module was shallow at the workflow level. Two subtly different version semantics coexisted with a stale contradictory comment, the GitHub fetch created its network adapter internally, and five failure kinds collapsed into None — callers had to know that meant unknown, not no update.

Change

• version_check/provider_updates.rs: check(CheckInput) -> Assessment (resolved binary path + --version output + npm output in; UpToDate / UpdateAvailable / Unknown{reason} out).
• version_check/app_releases.rs: check(current, Result<&[u8], FetchError>) -> Assessment — fetch is a separate adapter, tests feed fixture bytes to the check itself.
• providers.rs shrinks to gather-inputs → check → apply state + notices; Unknown handling is one explicit match (no notice, update_available = false, display fields preserved).
• Version-semantics finding: the provider's missing-patch→0 behavior is deliberate compatibility (the adjacent 'reject two-part' comment contradicted the code since inception) — both semantics kept, each documented in its own module.

Wins — tests reach the check, not just parsers (9 assessment tests); unknown is a first-class result; one release domain per module.

Behavior unchanged: same notices, same update rows, same silence on network failure — now as an explicit runtime choice.

Gates: fmt / clippy -D warnings / workspace tests — green (one unrelated concurrent-test interference on the shared ~/.tcode/worktrees dir was diagnosed separately and is being fixed in the worktree-lifecycle PR).

🤖 Generated with Claude Code