fix: broken E2E harness baseline on master (kimi-code guidance test) #2098

Description

The Validate harness and Compose definitions job (workflow: e2e-harness.yml) fails make harness-check on a clean checkout of current master (e4efb39b):

FAILED e2e/bub/tests/test_harbor_job_config.py::test_kimi_code_on_arm_installs_the_plugin_as_setup_does[authenticated]
FAILED e2e/bub/tests/test_harbor_job_config.py::test_kimi_code_on_arm_installs_the_plugin_as_setup_does[unauthenticated]

Assertion: _kimi_installation(harness) == _kimi_installation(tmp_path / "setup-home") — the harness-installed
plugins/managed/powercontext/skills/powercontext-project-context/SKILL.md does not match the one produced by
powercontext setup kimi-code (content diff inside the generated guidance file; 12 other mounted files are identical).

Why this matters

Every PR that touches anything matched by the harness workflow currently shows a red validate check even when the PR is unrelated, because the baseline is red. Recent master push runs:

• https://github.com/oceanbase/powercontext/actions/runs/39234099759 (failure, 36s)
• https://github.com/oceanbase/powercontext/actions/runs/39232839694 (failure)
• https://github.com/oceanbase/powercontext/actions/runs/39232604166 (failure)

Observed on PR #2093 as well; the PR itself does not touch kimi-code, guidance templates, or the harness.

Reproduction

git checkout e4efb39b
make harness-sync
uv run --project e2e/bub python -m pytest e2e/bub/tests/test_harbor_job_config.py -q
# 2 failed, 106 passed on macOS (arm, Python 3.12 via e2e/bub/.venv)

The failure content-diffs at
integrations/kimi-code/plugins/powercontext/skills/powercontext-project-context/.powercontext-guidance.json's pinned
legacy_sha256 of SKILL.md versus what the two install paths render, so either the rendered template or the pinned hash drifted. Note the same mismatch could be platform/locale-dependent if any template text is machine-dependent, which may explain why some environments pass.