The leak check compared only source skills, so a skill the user keeps in ~/.claude/skills (which merge mode preserves) passed as harmless, although OpenCode loads it. Read the writer's folder too and count entries sync did not put there and OpenCode's own folder lacks. The wording no longer says these skills are left out by filters, and the suggestion asks to sync them to opencode, which fits both causes. Refs #458
4.6 KiB
CLI E2E Runbook: OpenCode Skill Discovery vs Doctor
Scope
Check skillshare doctor's cross_target_discovery result against what a real
OpenCode loads when claude and opencode are both targets (#458). OpenCode
also reads ~/.claude/skills and ~/.agents/skills, and keeps one skill per
name across them.
- A skill excluded from
opencodestill reaches OpenCode through~/.claude/skills, and doctor names that skill. OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1stops OpenCode reading~/.claude/skills; with the variable set, doctor passes and notes the skip.- A skill synced to both folders loads once, and doctor passes.
opencode debug skill lists loaded skills without a login or model call.
Which copy of a duplicated name wins is not fixed, so the steps check counts,
not the winning folder.
Environment
Run inside the devcontainer with network access (step 1 installs opencode-ai
from npm into /tmp/opencode-skills). Helper: scripts/opencode/skill-discovery.sh.
Run mdproof directly, not inside ssenv enter: mdproof gives the runbook its
own HOME, while ssenv's XDG_CONFIG_HOME would still point skillshare and
OpenCode at the ssenv folder.
CONTAINER=$(docker compose -f .devcontainer/docker-compose.yml ps -q skillshare-devcontainer)
docker exec "$CONTAINER" /workspace/.devcontainer/ensure-mdproof.sh
docker exec "$CONTAINER" \
mdproof --report json /workspace/ai_docs/tests/opencode_skill_discovery_runbook.md
Steps
Step 1: Build and install OpenCode
cd /workspace && make build >/dev/null
bash /workspace/scripts/opencode/skill-discovery.sh install
Expected:
- exit_code: 0
- regex: opencode \d+.
Step 2: OpenCode loads a skill excluded from it through ~/.claude/skills
S=/workspace/scripts/opencode/skill-discovery.sh
bash "$S" fixture filtered
bash "$S" loaded
Expected:
- exit_code: 0
- jq: map(select(.name == "claude-only")) == [{"name": "claude-only", "from": "~/.claude/skills"}]
- jq: map(select(.name == "shared")) | length == 1
Step 3: Doctor names the leaked skill and the variable
bash /workspace/scripts/opencode/skill-discovery.sh fixture filtered
ss doctor -g --json | jq '.checks[] | select(.name == "cross_target_discovery")'
Expected:
- exit_code: 0
- jq: .status == "warning"
- jq: .details | any(test("loads skills missing from its own folder: claude-only$"))
- jq: .suggestions | any(test("OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1"))
Step 4: The variable hides the skill from OpenCode and passes doctor
S=/workspace/scripts/opencode/skill-discovery.sh
bash "$S" fixture filtered
L=$(bash "$S" loaded OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1)
D=$(OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1 ss doctor -g --json | jq '.checks[] | select(.name == "cross_target_discovery")')
jq -n --argjson loaded "$L" --argjson doctor "$D" '{loaded: $loaded, doctor: $doctor}'
Expected:
- exit_code: 0
- jq: .loaded | map(.name) == ["shared"]
- jq: .doctor.status == "pass"
- jq: .doctor.details | any(test("opencode skips .*\.claude/skills: OPENCODE_DISABLE_CLAUDE_CODE_SKILLS is set"))
Step 5: Skills synced to both folders load once and pass doctor
S=/workspace/scripts/opencode/skill-discovery.sh
bash "$S" fixture all
L=$(bash "$S" loaded)
D=$(ss doctor -g --json | jq '.checks[] | select(.name == "cross_target_discovery")')
jq -n --argjson loaded "$L" --argjson doctor "$D" '{loaded: $loaded, doctor: $doctor}'
Expected:
- exit_code: 0
- jq: .loaded | map(.name) == ["claude-only", "shared"]
- jq: .doctor.status == "pass"
Step 6: A skill the user keeps in ~/.claude/skills reaches OpenCode and doctor names it
S=/workspace/scripts/opencode/skill-discovery.sh
bash "$S" fixture all
mkdir -p ~/.claude/skills/my-local
printf -- '---\nname: my-local\ndescription: kept by hand in claude\n---\nbody\n' > ~/.claude/skills/my-local/SKILL.md
L=$(bash "$S" loaded)
D=$(ss doctor -g --json | jq '.checks[] | select(.name == "cross_target_discovery")')
jq -n --argjson loaded "$L" --argjson doctor "$D" '{loaded: $loaded, doctor: $doctor}'
Expected:
- exit_code: 0
- jq: .loaded | map(select(.name == "my-local")) == [{"name": "my-local", "from": "~/.claude/skills"}]
- jq: .doctor.status == "warning"
- jq: .doctor.details | any(test("loads skills missing from its own folder: my-local$"))
Step 7: Clean up
bash /workspace/scripts/opencode/skill-discovery.sh clean
test ! -e /tmp/opencode-skills && echo "opencode removed"
Expected:
- exit_code: 0
- opencode removed
Pass Criteria
Steps 1 to 7 pass: doctor warns exactly when OpenCode loads a skill sync keeps out of it, and passes when the variable is set or every skill reaches both folders.