test(ci): wire the 147 Idris2 + 186 Zig integration tests into CI; clear RSR template drift

[hyperpolymath]

Summary

Follow-up hygiene from the repo-wide audit (the same audit that produced the #82 rewrite). Two independent wins, each verified with the real toolchain, not asserted:

• (a) Run the tests that already exist but never ran in CI. The 147 Idris2 ABI tests were only typechecked; the 186-case Zig integration suite was never wired into build.zig and, it turns out, never compiled. Both now build and pass, and both run in CI.
• (b) Clear three RSR template-drift bugs the audit flagged (a live {{…}} placeholder, a stale repo identity, and a version drift).

Toolchain used locally: idris2 0.8.0 (Chez, matching abi-typecheck.yml) + zig 0.15.2 + GTK/WebKit (matching e2e.yml).

Changes

(a) CI test wiring — commit test(ci): …

• abi-typecheck.yml — after typechecking shell+groove, idris2 --build + run gossamer-abi-tests → 147 passed, 0 failed. Path filters extended to tests/idris2/** and gossamer-abi-tests.ipkg.
• e2e.yml + src/interface/ffi/build.zig — new test-integration step, run in the E2E job → 186 passed. Making test/integration_test.zig compile required:
  • main.zig re-exports its FFI submodules (csp/filesystem/ssg/groove/plugin/clipboard/dialog/tray) so a test/-rooted module can reach them through the gossamer module (Zig 0.15 forbids ../src imports across a module root — the reason it was never wired);
  • completing the export fn → pub export fn migration on the C-ABI surface (identical C symbols, now Zig-visible) so the test can call them;
  • fixing three latent bugs the never-built suite hid: gossamer_notify returns u32 not Result; std.mem.zeroes(GossamerHandle) is rejected by Zig 0.15 (byte-zero the handle instead); the Result enum has 12 variants now, not 11 — and Zig main.zig and Idris2 Types.idr agree (verified: guard_locked=11 ↔ GuardLocked=11).
• Regression-checked: default test, test-display, test-android, and the libgossamer.so artifact all still build/pass.

(b) RSR drift — commit chore(rsr): …

• bot_directives/methodology.a2ml — fill the {{PROJECT_UNIQUE_STRENGTH}} placeholder (the file's own reject-if-contains rule flags unresolved {{…}}).
• ai/AI.a2ml — repo identity claimed to be rsr-template-repo; corrected to gossamer (also self-flagged by the same reject rule).
• .well-known/groove/manifest.json — service_version 0.1.0 → 0.3.1 (matches the canonical root 0-AI-MANIFEST.a2ml).
• .editorconfig — stale RSR-template-repo header → gossamer.

RSR Quality Checklist

Required

• Tests pass (idris2 147/0; zig integration 186/0; default/display/android green)
• Code is formatted (Zig edits are pub prefixes + submodule re-exports; consistent with house style)
• Linter is clean (no new warnings)
• No banned language patterns (Idris2 / Zig / YAML / a2ml / JSON only)
• No unsafe blocks without // SAFETY: comments (none added)
• No banned functions (believe_me, unsafeCoerce, Obj.magic, Admitted, sorry)
• SPDX headers unchanged on all touched source files
• No secrets, credentials, or .env files included

As Applicable

• ABI/FFI changes validated (src/interface/abi/ ↔ src/interface/ffi/ — the 186 integration tests now enforce result-code/ABI agreement in CI)
• CHANGELOG — intentionally omitted (CI-plumbing + manifest hygiene; happy to add an entry if you'd prefer)

Testing

# abi-typecheck.yml flow, locally (idris2 0.8.0):
guard PASS · shell GREEN(11) · groove GREEN(4) · gossamer-abi-tests → 147 passed, 0 failed

# e2e.yml step, locally (zig 0.15.2 + GTK/WebKit):
zig build test-integration → 186 passed
zig build test / test-display / test-android → all green
zig build → libgossamer.so.0.3.0 (no regression)

Companion to the #82 rewrite (done separately, no code change). Deliberately scoped to CI plumbing + manifest hygiene — no product-behaviour change.

🤖 Generated with Claude Code

https://claude.ai/code/session_013Zg1kEsjSRpcf1wFBZyeNj

---

Generated by Claude Code