Emit JSONL from every assistant in the test harness (#41)

* test-assistants.sh: emit JSONL from Codex and Copilot CLI

Codex (exec --json) and Copilot CLI (--output-format json) now stream JSONL like
Claude Code and Cursor, so four of five assistants share one output shape through
unwrap_log(). Verified with a smoke run: both logs are JSON and their reports
still parse (report fields and COMMANDS extracted, both PASS). Antigravity is left
on plain text until its quota allows a verification run.

* test-assistants.sh: emit stream-json from Antigravity too

Adds --output-format stream-json to the agy launch, matching Claude Code and
Cursor, so all five assistants share one output shape through unwrap_log().
NOT yet verified: Antigravity's weekly quota is exhausted, so its report can't
be exercised in a run. The flag choice is low-risk — stream-json is the same
format Claude Code and Cursor already emit and parse — but this leg stays
unverified until a run can confirm it. PR remains draft until then.

* README: all five assistants verified end-to-end

Antigravity CLI's end-to-end path is now verified against a live tenant (it authored a workflow from the canonical prompt and imported it), so the verification note no longer needs to single it out as unverified.

Author: mraible

README.md

@@ -26,7 +26,7 @@ AI coding assistant skills for building [CrowdStrike Falcon Fusion](https://www.
 | Cursor | `/add-plugin crowdstrike-falcon-fusion` | [Cursor](https://cursor.com/marketplace/crowdstrike) (pending) |
 | Antigravity CLI | `agy plugin install https://github.com/CrowdStrike/fusion-skills` | [Google](https://antigravity.google/docs/plugins) |
 
-In a live-tenant run, Claude Code, Codex, Copilot CLI, and Cursor each authored a valid workflow from the example prompt below and imported it to the tenant. Antigravity CLI loads the skills, but its weekly quota was exhausted before an end-to-end run.
+In a live-tenant run, all five assistants (Claude Code, Codex, Copilot CLI, Cursor, and Antigravity CLI) each authored a valid workflow from the example prompt below and imported it to the tenant.
 
 These skills follow the [Agent Plugins](https://agent-plugins.org) format, with a root `plugin.json` and a Codex `.codex-plugin/plugin.json` so the non-Claude assistants can discover them.
 

test-assistants.sh

@@ -455,10 +455,10 @@ unlink_repo_skills() {
 # repo's symlinks present there. Cursor's CLI binary is `agent` (also installed as `cursor-agent`; both are the same executable).
 ASSISTANTS=(
   "Claude Code|claude|--plugin-dir|-p %%PROMPT%% --plugin-dir $REPO --dangerously-skip-permissions --verbose --output-format stream-json"
-  "Codex|codex|~/.agents/skills|exec %%PROMPT%% --skip-git-repo-check"
-  "Copilot CLI|copilot|--plugin-dir|-p %%PROMPT%% --plugin-dir $REPO --allow-all"
+  "Codex|codex|~/.agents/skills|exec %%PROMPT%% --skip-git-repo-check --json"
+  "Copilot CLI|copilot|--plugin-dir|-p %%PROMPT%% --plugin-dir $REPO --allow-all --output-format json"
   "Cursor|agent|--plugin-dir|-p %%PROMPT%% --plugin-dir $REPO --force --trust --output-format stream-json"
-  "Antigravity CLI|agy|~/.agents/skills|-p %%PROMPT%% --dangerously-skip-permissions"
+  "Antigravity CLI|agy|~/.agents/skills|-p %%PROMPT%% --dangerously-skip-permissions --output-format stream-json"
 )
 
 want() {