feat(hogli): add ci:insights adapter (with Mendral concrete class)

[rnegron]

Problem

Mendral (our CI-insights backend) holds aggregated cross-run intelligence — recurring flakes, occurrence counts, confidence, proposed/merged fixes — that no single CI run can show. Today it lives only in Slack and the web UI, so engineers (and agents) rarely consult it at the moment they're actually debugging a red build.

Changes

• New hogli ci:insights command — an adapter over the CI-insights backend. Verbs: ci:insights (digest for the current repo + branch, default), search, view [--json], plan (prints the recommended remediation plan; does not apply it). All Mendral-specific knowledge is quarantined in one MendralBackend class. Lives in the hogli-commands extension layer; core untouched.
• debugging-ci-failures skill gains a "Check existing CI insights first" step (between Classification and Local reproduction) so agents query an existing hypothesis before reproducing from scratch.

How did you test this code?

I'm an agent; I did not perform manual QA beyond the runs below. Automated + live checks I actually ran:

• test_ci_insights.py — 16 passing (preflight, verb→backend mapping, exit-code propagation, recommended-action selection).
• hogli ci:insights / search / plan <id> against the live backend; plan correctly selected a proposed recommended action over a merged one; bad id exits non-zero.
• ruff check/format clean; hogli lint:skills + build:skills pass with no file drift.

Publish to changelog?

No.

🤖 Agent context

Authored with Claude Code

[greptile-apps]

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
tools/hogli-commands/hogli_commands/tests/test_ci_insights.py:45-53
The parametrize table covers `view --json` but omits the analogous `search --json` case, leaving the `--json` flag on `search` untested despite the project preference for parameterised tests covering symmetric behaviour.

```suggestion
@pytest.mark.parametrize(
    "argv, expected",
    [
        ([], ("mendral", "here")),
        (["search", "flaky timeout"], ("mendral", "insight", "search", "flaky timeout")),
        (["search", "flaky timeout", "--json"], ("mendral", "insight", "search", "flaky timeout", "--json")),
        (["view", "01ABC"], ("mendral", "insight", "view", "01ABC")),
        (["view", "01ABC", "--json"], ("mendral", "insight", "view", "01ABC", "--json")),
    ],
)
```

### Issue 2 of 2
tools/hogli-commands/hogli_commands/tests/test_ci_insights.py:106
The payload `'{"actions": [null]}'` is valid JSON, so it never reaches the `json.JSONDecodeError` handler — the null entry is silently filtered and the code lands on the "No remediation plan available" path, which is already the focus of `test_plan_errors_when_no_actions`. Keeping it here conflates two distinct failure modes; moving it there as a second parametrize case keeps the intent explicit.

```suggestion
@pytest.mark.parametrize("payload", ["", "not json{"])
```

_{Reviews (1): Last reviewed commit: "fix(hogli): harden ci:insights edge case..." | Re-trigger Greptile}

[stamphog]

The hex-security-app bot raised a valid, current-head, unresolved argument-injection concern: query and insight_id are passed directly as positional elements in the subprocess.run args list with no -- option terminator, so a value like --config=... or --token=... would be interpreted as a flag by the mendral CLI rather than as a literal argument. The fix is a one-line change (place "--" before each user-supplied positional in search, view, and plan). Per review policy, substantive bot comments with valid concerns that remain unaddressed are a REFUSE.

[stamphog]

The hex-security-app bot flagged argument injection in search, view, and plan — user-supplied query and insight_id are passed as positional args to subprocess.run without a -- terminator, so a value starting with - would be interpreted as a flag by the mendral CLI. The thread was marked resolved but the code was not changed, and the stamphog aggregator top-level review (current head) explicitly calls this out as still unaddressed. The fix is trivial — add "--" before each user-supplied positional — and should be applied before auto-approving.

[webjunkie]

Clever 😏

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-05-28 13:58 UTC	Run
prod-us	✅ Deployed	2026-05-28 14:15 UTC	Run
prod-eu	✅ Deployed	2026-05-28 14:14 UTC	Run