ainb skill-manager · test coverage
Test coverage matrix & recorded lifecycle demos
38 of 44 skill-manager features (86%) carry at least one automated assertion
— 28 via CLI tests, 19 via tmux tripwires (9 on both axes); 6 features remain GAP on both axes (all TUI-only nav / launcher / legend surfaces). Two full lifecycles (CLI + TUI) are captured end-to-end as vhs recordings below.
44
Features tracked
28
CLI test covered
19
tmux tripwire covered
9
Both axes covered
6
GAP on both axes
Coverage matrix
Both CLI + tmux · Partial one axis · GAP neither. Olive cells = covered, clay cells = GAP.
| Feature | CLI test | tmux tripwire | Status |
|---|---|---|---|
| TUI Screen Layout Three-pane chrome (Sources/Units/Detail) asserted via screen_opens + sandbox_e2e. No CLI surface — TUI-only. Help-bar keybind legend not separately asserted as a string. | GAP | tripwire_core_skill_manager_screen_opens.rs | Partial |
| Discovery Banner First-open overlay + [Enter] import asserted live. [d] details-toggle and [s] skip-with-persistence are NOT asserted (only [Enter] import path is). | GAP | tripwire_core_skill_manager_discovery_banner.rs | Partial |
| [i] Add Source Keybind keys_wired presses [i], types a colon-bearing URI, Enter, asserts source lands in manifest.yaml. CLI side via source_cli_tests (ainb source add). | source_cli_tests.rs | tripwire_core_skill_manager_keys_wired.rs | Both |
| [u] Update Keybind CLI path (ainb skill update) well covered. No tmux tripwire presses [u] in-screen — only the underlying CLI flow is tested. | skill_update_tests.rs | GAP | Partial |
| [c] Check Keybind keys_wired doc-comment mentions [c] but only [i] and [/] are actually asserted. No tmux tripwire presses [c] to re-trigger a drift scan. CLI ainb skill check covered. | skill_check_tests.rs | GAP | Partial |
| [r] Remove Keybind CLI path (ainb skill remove) covered. No tmux tripwire presses [r] in-screen. | skill_remove_tests.rs | GAP | Partial |
| [s] Sync Keybind (No Conflict) s_routes_to_sync_live presses [s] on a non-shadowed unit, asserts SkillManagerSync + 'sync: <unit>' notification. | skill_sync_tests.rs | tripwire_core_skill_manager_s_routes_to_sync_live.rs | Both |
| [s] Conflict Flip Keybind conflict_flip presses [s] on a shadowed_by pair, asserts manifest edge flips + idempotent roundtrip + no-op on non-conflict. No CLI equivalent for the flip routing. | GAP | tripwire_core_skill_manager_conflict_flip.rs | Partial |
| [/] Search Keybind keys_wired presses [/], filters Units table case-insensitively, guards the ':' palette regression. search_cli_tests covers the CLI search surface. | search_cli_tests.rs | tripwire_core_skill_manager_keys_wired.rs | Both |
| [m] Refresh Discovery Keybind GAP both axes. All tripwires use [m] from HOME to OPEN the screen, never as the in-screen refresh-discovery key that clears .discovery-skipped. The marker-clearing/force-rebanner behavior is unasserted. | GAP | GAP | GAP |
| Arrow/j/k Navigation GAP. No tripwire presses arrow/j/k to move selection and recompute Detail. TUI-only feature with no nav tripwire and no CLI surface. | GAP | GAP | GAP |
| Home/g / End/G Navigation GAP both axes. First/last jump nav is entirely unasserted in tmux and has no CLI surface. | GAP | GAP | GAP |
| Discovery Class-A Walker class_a_walker_tests (CLI crate) covers marketplace cache scan + URI synthesis. No dedicated tmux tripwire (banner test exercises it indirectly via reconciler output). | class_a_walker_tests.rs | GAP | Partial |
| Discovery Class-C Walker discover_tests covers orphan-unit scan across adapter homes. No dedicated tmux tripwire; only the merged reconciler output surfaces in the banner test. | discover_tests.rs | GAP | Partial |
| Discovery Reconciler discover_tests exercises Class-A+Class-C merge / orphan-wins / shadowed_by. discovery_banner asserts the banner counts equal reconciler output live. | discover_tests.rs | tripwire_core_skill_manager_discovery_banner.rs | Both |
| Drift Detection Column drift_column (TestBackend) renders all glyph/colour combos incl '…' placeholder. drift_glyph_live also covers ⚠ live. No CLI column rendering (CLI check is tabular text). | GAP | tripwire_core_skill_manager_drift_column.rs | Partial |
| Background Drift Poll drift_background_poll asserts GoToSkillManager spawns poll, lands in drift_cache on next tick, coalesces in-flight. Mocked backend. No CLI surface. | GAP | tripwire_core_skill_manager_drift_background_poll.rs | Partial |
| Drift GitLsRemoteBackend drift_tests_integration covers InSync/Outdated/argv-smuggle/GIT_TERMINAL_PROMPT against real bare repo. drift_glyph_live proves it end-to-end in the live binary. | drift_tests_integration.rs | tripwire_core_skill_manager_drift_glyph_live.rs | Both |
| Skill Usage Tracking lockfile_tests covers schema-v2 LockedUnit.usage round-trip. usage_renders (TestBackend) + usage_renders_live assert Detail-pane rendering. | lockfile_tests.rs | tripwire_core_skill_manager_usage_renders.rs | Both |
| ainb skill usage CLI skill_usage_tests seeds fake session-log JSONL + asserts lockfile usage refresh. detect_invocations covers the parser. No tmux surface for the CLI command itself. | skill_usage_tests.rs | GAP | Partial |
| Detail Pane Render usage_renders + usage_renders_live assert URI/usage/time-ago in Detail pane. format_time_ago fallback covered. TUI-only, no CLI. | GAP | tripwire_core_skill_manager_usage_renders.rs | Partial |
| [s] Sync to Home Executor sync_to_home_tests (core) verifies bytes land at mapped home path idempotently via ContentFetcher. No tmux tripwire isolates the to-home executor; s_routes_to_sync only asserts routing. | sync_to_home_tests.rs | GAP | Partial |
| [s] Sync to Repo Executor sync_to_repo_tests + skill_sync_roundtrip_tests cover home→repo copy/commit/push idempotency + argv hardening. No tmux tripwire for the to-repo executor. | sync_to_repo_tests.rs | GAP | Partial |
| Sync SyncPlanner sync_tests covers plan_sync ToRepo/ToHome/NoOp decision logic (pure fn). No TUI tripwire; not a user-visible surface on its own. | sync_tests.rs | GAP | Partial |
| ainb skill sync CLI skill_sync_roundtrip_tests + skill_sync_tests cover --to-repo/--to-home/--dry-run end-to-end against git sources. No tmux surface for the CLI command. | skill_sync_roundtrip_tests.rs | GAP | Partial |
| Auto-infer target_layout auto_infer_gh_layout_tests covers gh:owner/repo fall-through to BOOTSTRAP_DEFAULT_MAPPINGS. mapping_tests covers resolve_pair. Pure-fn, no TUI surface. | auto_infer_gh_layout_tests.rs | GAP | Partial |
| ainb skill check CLI skill_check_tests covers tabular output, --json variant, per-source scoping with MockBackend. No tmux surface for the CLI command. | skill_check_tests.rs | GAP | Partial |
| ainb skill promote Command promote_tests drives full local:->git: flow against hermetic bare repo (file://): bare-repo receives skills/, manifest URI rewritten, lockfile sha256. No TUI/tmux surface — promote is CLI-only. | promote_tests.rs | GAP | Partial |
| Promote Pre-flight Checks Covered within promote_tests (gh auth / manifest-has-unit / local: / defaultBranchRef resolution, fail-fast before writes). No tmux surface. | promote_tests.rs | GAP | Partial |
| Promote Failure Modes promote_tests is the only home; verify it asserts each failure branch (auth fail, push fail leaves manifest unrewritten, network down). No tmux surface. | promote_tests.rs | GAP | Partial |
| Promote Cache Persistence Clone-once/reuse-via-pull cache layout exercised inside promote_tests. No dedicated persistence-across-runs assertion called out; no tmux surface. | promote_tests.rs | GAP | Partial |
| Sandbox Test Fixture (Minimal Tier) sandbox_fixture_smoke asserts minimal-tier build + seeded skills. sandbox_loads renders it via TestBackend. | sandbox_fixture_smoke.rs | tripwire_core_skill_manager_sandbox_loads.rs | Both |
| Sandbox Test Fixture (Full Tier) sandbox_fixture_smoke covers both tiers building correctly. sandbox_e2e exercises the full-tier fixture in the live binary. | sandbox_fixture_smoke.rs | tripwire_core_skill_manager_sandbox_e2e.rs | Both |
| Sandbox Rust API sandbox_fixture_smoke covers build_skill_manager_sandbox / env_vars round-trip / refuse-$HOME / idempotent rebuild. API-level, no TUI surface. | sandbox_fixture_smoke.rs | GAP | Partial |
| Sandbox Bash Launcher GAP both axes. scripts/skill-manager-sandbox.sh (up/down, safety guards, .ainb-sandbox-marker) exists but has no automated test — exercised only by hand. | GAP | GAP | GAP |
| Sandbox Just Recipes GAP both axes. skill-manager.just module (up/tui/cli/inspect/down) exists but is untested in CI; only manually invoked. | GAP | GAP | GAP |
| Dotdir Strip Fix sync_strip_tool_dotdir_test is the dedicated regression for strip_tool_dotdir() in apply_to_home/repo. Pure-fn, no TUI surface. | sync_strip_tool_dotdir_test.rs | GAP | Partial |
| Per-Source Advisory Lock concurrent_sync_test verifies the .ainb-sync.lock advisory lock prevents git-index contention across units. No TUI surface. | concurrent_sync_test.rs | GAP | Partial |
| Input State Modal keys_wired exercises InputState for both AddSource ([i]) and Search ([/]) — keystroke-to-buffer until Enter/Esc, centered overlay. No CLI surface. | GAP | tripwire_core_skill_manager_keys_wired.rs | Partial |
| Selected Unit Conflict Peer Check conflict_flip + s_routes_to_sync_live together prove the manifest-read routing (sync vs flip) based on shadowed_by peer. No CLI surface. | GAP | tripwire_core_skill_manager_conflict_flip.rs | Partial |
| Visible Indices Singleton keys_wired [/] filter proves visible_indices() drives render+filter agreement (Units empties, Detail keeps name). Nav-side of the singleton (selection movement) is NOT separately asserted. | GAP | tripwire_core_skill_manager_keys_wired.rs | Partial |
| CLI run_skill_cli / run_source_cli Helpers In-process dispatch + notification capture proven by keys_wired ([i] -> manifest write) and s_routes_to_sync_live (notification surfaces). source_cli_tests covers the dispatch path directly. | source_cli_tests.rs | tripwire_core_skill_manager_keys_wired.rs | Both |
| SkillsScreenData Live Data Binding screen_opens asserts Manifest+Lockfile+UsageCache populate Sources/Units/Detail on GoToSkillManager. sandbox_loads complements in-process. No CLI surface. | GAP | tripwire_core_skill_manager_screen_opens.rs | Partial |
| Help Bar Keybinds GAP both axes. The advertised legend string ([i][u][c][r][s][/][?][q] with action names + gold/muted styling) is never asserted as a rendered string; individual key wiring is tested but not the legend itself. | GAP | GAP | GAP |
Gaps (14)
- [m] Refresh Discovery Keybind — GAP on BOTH axes. Every tripwire uses [m] from HOME to OPEN the screen; none presses [m] in-screen to clear the .discovery-skipped marker and force the banner to re-appear past a prior [s] skip. The 'empty-state hint finally works' behavior (the actual point of this feature, per recent commit eced97e/5a5cd17) is unverified end-to-end.
- Arrow/j/k Navigation — GAP on BOTH axes. No tripwire presses arrow/j/k to move selection through visible units, verify wrap-at-ends, or confirm Detail-pane recompute on move. TUI-only with zero coverage despite being core nav.
- Home/g / End/G Navigation — GAP on BOTH axes. First/last jump (Home/g, End/G) and its wrap consistency is entirely unasserted.
- [u] Update Keybind — tmux GAP. CLI ainb skill update is covered (skill_update_tests.rs) but no tripwire presses [u] in-screen to prove the keybind dispatches the in-process update + surfaces a result notification.
- [c] Check Keybind — tmux GAP. keys_wired.rs doc-comment claims [c] is wired but only [i] and [/] are actually asserted. No tripwire presses [c] to re-trigger the drift scan and confirm the status column refreshes.
- [r] Remove Keybind — tmux GAP. CLI ainb skill remove covered (skill_remove_tests.rs) but no tripwire presses [r] in-screen to prove the selected unit is uninstalled and the panels refresh.
- Help Bar Keybinds (legend render) — GAP on BOTH axes. Individual key wiring is tested piecemeal, but the rendered legend string itself ([i]nstall [u]pdate [c]heck [r]emove [s]ync [/]search [?]help [q]uit with gold-key/muted-desc styling) is never asserted as a buffer substring, so a dropped/renamed legend entry would pass silently.
- Sandbox Bash Launcher — GAP on BOTH axes. scripts/skill-manager-sandbox.sh exists (up/down, env exports, $HOME/'/' safety guards, .ainb-sandbox-marker teardown protection) but has no automated test; its safety guards are the riskiest untested surface.
- Sandbox Just Recipes — GAP on BOTH axes. skill-manager.just (up/tui/cli/inspect/down, AINB_SANDBOX_ROOT override) exists but is exercised only by hand.
- Discovery Banner secondary keys — partial GAP. discovery_banner.rs asserts [Enter] import but NOT [d] details-toggle or [s] skip-with-persistence-marker; the skip marker write (prerequisite for the [m]-refresh feature above) is never asserted.
- Discovery Class-A / Class-C walkers — tmux GAP. Both have CLI tests (class_a_walker_tests.rs, discover_tests.rs) but neither walker has a dedicated tmux tripwire; they are only exercised indirectly through the merged reconciler output in discovery_banner.rs.
- Sync to Home / Sync to Repo executors — tmux GAP. Both have strong core pure-fn tests (sync_to_home_tests.rs, sync_to_repo_tests.rs) but no tmux tripwire isolates the executor reached via the [s] keybind; s_routes_to_sync_live only asserts the routing event fires, not that bytes actually moved.
- Promote family ([s]/CLI) — tmux GAP across all four promote features (command, pre-flight, failure-modes, cache-persistence). Covered only by promote_tests.rs at CLI level; promote has no TUI keybind, so this is expected, but the failure-mode branches (push fails -> manifest NOT rewritten) should be individually confirmed inside promote_tests.rs.
- Visible Indices Singleton — partial. keys_wired.rs proves the render+filter side ([/] empties Units, Detail keeps name) but the selection-movement side of the singleton (highlighted row tracking through nav) is unverified because there is no arrow/j/k tripwire.
Recorded lifecycle demos
CLI lifecycle
source add → source list → skill install (9 tools) → skill check / --json → edit deployed SKILL.md → skill check --json → skill sync --to-repo --dry-run → skill remove --yes → ls confirms removal. Recorded on a fresh minimal-tier sandbox at /tmp/sm-rec-cli-lifecycle; binary ainb 1.2.1 built under real HOME.
recordings/cli-lifecycle.gif
TUI lifecycle
m (open Skill Manager) → Down ×2 (navigate Units, recompute Detail) → / commit (search filter) → Esc (clear) → s (sync selected unit) → q (return to HomeScreen). Recorded on a full-tier sandbox at /tmp/sm-rec-tui-lifecycle; Dracula theme, 1400×800, ainb 1.2.1.
recordings/tui-lifecycle.gif
Generated from the skill-manager coverage audit. 2 of 2 recordings embedded as GIF. GIFs were produced with vhs against the real ainb 1.2.1 binary in isolated sandboxes (/tmp/sm-rec-cli-lifecycle, /tmp/sm-rec-tui-lifecycle), torn down after capture. No fabricated data — every test name above maps to a real file on feat/skill-manager.