Repository navigation
Build a reproducible Raycast command compatibility catalog and test batches #72
Description
Part of #1 and #7. Prioritized by the user after the first touch/readability slice. Build a reproducible catalog and expand compatibility in bounded, reviewable batches; do not advertise all Raycast extensions as tested.
Starting evidence
The open-source raycast/extensions snapshot 11a912d89b576ca066e9ee42e75b8dd2e8aa0f5c contains 3,491 extension directories, obtained from a nontruncated direct Git tree. This is a repository-directory inventory, not the entire Raycast Store, a package-manifest count, or an execution result. The initial inventory performs no new functional verifications. Existing bounded coverage includes UUID Generator, OKLCH Color Converter and Palette Colors, with command/build/source-specific limits in PARITY.md and the #66 report.
Acceptance criteria
- Add a deterministic, resumable inventory tool that pins upstream commit/tree identities and records counts, truncation/completeness, manifest retrieval failures and scope. Keep open-source repository and live Store inventories distinct; retain archived snapshots and compare additions/removals rather than silently changing the denominator.
- Index extensions and individual commands, declared desktop platforms, modes, version/source identities and relevant requirements. Static dependency/API inspection supplies screening signals only; it cannot promote a command to working or label an untested command as execution-failed.
- Use explicit statuses: untested, tested-working-for-listed-workflow, tested-partial, observed-failed, platform-unsupported-with-specific-reason, and requires-user-configuration/authentication. Keep installation, startup/rendering, interaction/action result, destination, persistence and lifecycle checks distinct. Display the exact tested workflow and environment.
- Define a representative first batch covering offline utility/List/Grid/Detail/Form/no-view/network/arguments/preferences/storage, and record exact tested bundle hashes/commands, Android/WebView/API, viewport and host build. Include actual native success/error/destination screenshots and controlled expected results; preserve original failures.
- Use isolated app/extension data, owned fixtures and bounded networking/cleanup. Never overwrite installed extensions or run commands against unrelated user data. Do not submit authenticated actions, spend money, send messages or connect personal accounts implicitly; record configuration-blocked flows honestly.
- Inventory unsupported desktop capabilities (AppleScript/process execution, macOS app/window integration, OAuth/native dependencies, menu-bar/background behavior and unsupported host/UI APIs). Show actionable, specific inline errors and create enhancement issues for safe Android host additions only when evidence justifies them.
- Add rate/size bounds, retries only for transient retrieval errors, explicit remaining-work state and reproducible command-line outputs. Full catalog discovery is not a claim of full functional coverage. A catalog unavailable/truncated error must never become zero extensions or 100% passing.
- Keep the matrix in source docs and link it from the extension UX when useful; no unverified compatibility badges. Per implementation PR: real app smoke/screenshots, appropriate local checks, two independent gpt-6.1-sol/high exact-head audits and an expected-head merge.
Primary references: https://github.com/raycast/extensions/tree/11a912d89b576ca066e9ee42e75b8dd2e8aa0f5c ; https://developers.raycast.com/information/manifest ; https://developers.raycast.com/information/lifecycle . Continue API26 recovery/input issues #68/#69 alongside this work, without claiming their compatibility is certified.
Activity
moikdev commentedon Nov 1, 2026
Discovery checkpoint for pinned raycast/extensions source 11a912d: the resumable prototype retrieved 3,491 package manifests and indexed 11,159 command declarations, with no retrieval/manifest errors. Modes: 7,507 view, 3,269 no-view, 383 menu-bar. This is the pinned open-source snapshot, not the entire live Store.
All 11,159 commands remain untested. No extension code, authenticated action, account connection or runtime compatibility promotion was performed. Static mode/platform/dependency declarations are screening data only.
All 3,491 cached payloads match their recorded SHA-256/Git-blob hashes; the 20 pilot manifests also match their independently retained expected hashes. A zero-request resume after the first 100 manifests produced a byte-identical catalog export. The collector is still an unpublished prototype; formal tooling checks, reviewed source documentation and command/bundle/environment-specific execution batches remain required. No compatibility badges or passing percentage should be derived from this discovery count.
The first product priority remains touch/readability #71; this read-only discovery work ran while its owned-emulator checks were in progress.
Add a pinned, explicitly untested Raycast command catalog (#73)
moikdev commentedon Nov 1, 2026
The discovery tooling is now merged in PR #73 (reviewed 86218d1, identical merge tree 66d5d94). Reproduction, pinned source/anchor inputs and validation receipts are in docs/EXTENSION-CATALOG.md and docs/validation/extensions/.
The stricter parser rejects duplicate JSON keys in monitor-mate and vat-calculator. The normalized checkpoint is 3,489 accepted manifests / 11,153 untested declarations, plus those two invalid manifests; all 3,491 directories were attempted. The earlier permissive total of 11,159 includes six excluded declarations. Discovery remains explicitly incomplete while those entries are invalid. Every exported command is untested, with zero runtime-tested commands.
Both independent gpt-6.1-sol/high final-head audits cleared the tooling after queue-cancellation and truncated-HTTP-read fixes. Matching regression tests fail on the previous collector and pass now; 18 collector cases pass normally and optimized, all 22 Python checks pass, and the hosted Android build passed. The retained Android smoke is clearly labeled and all 61 production/build source hashes match its original runtime. This is discovery preparation, not closure of the runtime-workflow acceptance criteria.
moikdev commentedon Nov 2, 2026
PR #76, head
8fd173baaadaf3e19ecc1cd04e9cb99d1211a141, publishes the first bounded Store workflow batch. Nine frozen bundles declare 34 commands; sixteen named commands were exercised. Eleven have completed listed workflows on both tablet/phone layouts, three remain partial and two have observed failures. Tablet History uses a focused same-APK pass after a SystemUI clipboard-helper interruption; the original broad untested row is preserved.The
eef1b8dprofile has 33 command observations, 70 raw capture sets and all 41 cleanup checks. A later collector opt-in repair has its own06d16edbinary/checkpoint: matched old missing-fixture failure → assumption skip, fresh configured History pass, two more raw capture sets and all 25 additional cleanup checks. Workflow/capture bodies and production/Python sources are unchanged; the older profile is not relabeled as a full rerun on the newer pair.Current local debug/test builds and 147 JVM cases pass; lint has 0 errors/43 warnings. Both 44-case Python runs pass with unchanged tooling. Report and per-command limits. Two fresh exact-head gpt-6.1-sol/high audits and the actual hosted head check remain merge gates.
Next: HTTP sandbox startup #75, native EmptyView/error rendering #6, explicit selected-text session input #19, and translation/service diagnosis. The pinned open-source inventory remains untested; these Store-version-specific outcomes do not change its denominator or close #72/#7.
moikdev commentedon Nov 2, 2026
The first workflow tooling/evidence slice merged in PR #76 as
10daace, with the exact audited8fd173btree, both independent sol/high audits, hosted Android and local/android gates passing. The report's scoped eleven completed workflows, three partial and two observed failures remain version-bound. Discovery inventory and the other bundle declarations remain untested. #72 stays open for further batches and the compatibility view; #75 is now the active runtime follow-up.moikdev commentedon Nov 4, 2026
Created #105 for the remaining user-facing compatibility view. It separates installed payload matching, historical recorded workflow outcomes, tested runtime/environment and unknown/error states. Complete payload identity (including assets/manifest) must support any match; titles, Store source metadata and installation are insufficient. A read-only native view and explicit evidence link should make existing results useful to touch users while retaining the precise status of untested commands and the incomplete pinned source inventory. Queue it after #104's verified merge; no new functional coverage is claimed here.
moikdev commentedon Nov 5, 2026
PR #106 merged as
8ff8f4d032ee351626fbfcbd6d5998e32e9577b4, closing #105. Root verified parentsc882c08a+ reviewed6ccb1e29and identical tree0f89f6fa455f5ead019690e5c54b1bb1b5a6360b. Both NEW independent full-change gpt-6.1-sol/high audits found no actionable findings. Actual exact-head Android run41944911428 and separate local/android58106906644 succeeded; merge used the expected-head SHA and the watch is disabled. The independent reviews are recorded in PR review5778372087 (COMMENT; GitHub disallowed self-approval). No source changed after review.Scoped evidence remains 44 native viewer tests /six cohorts /116 restoration checks, four separate intended negatives /zero passes /38 checks, 99 actual raw capture sets and full source/APK/publication identities. The app's historical catalog covers16 observations for14 of34 declarations across9 source snapshots;20 remain Not tested in this selected catalog. This is no new command-execution, whole-Store, physical/OEM/accessibility, production-signing or #89 clearance.