Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →stickerdaniel/2-repro-issue
Reproduce a single LinkedIn-MCP issue locally on the current branch against the real authenticated LinkedIn session at ~/.linkedin-mcp/profile/, using the MCP streamable-http server. Captures the exact failure mode (tool output, error, missing data) and maps it back to the scraper code path.
Passed 8 of 14 test cases with the skill, and 5 of 14 without it.
and 8 more in the full report
8 fixed6 passed either way8 still failing3 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
A newer attempt on gemini-3.8-flash on 10/18/2026 did not produce a publishable comparison (not comparable). The snapshot above is the previously measured run.
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
Before spawning a background MCP server process for issue reproduction, what git command should be run to inspect if the workspace has uncommitted changes, and what action should be taken if it is dirty?
| Checked | Without skill | With skill |
|---|---|---|
| Runs git status --porcelain to check for uncommitted changes. | Missed | Passed |
| Asks for user confirmation if the workspace is dirty. | Missed | Passed |
The instructions your model reads when the skill loads.
Evaluate the reporter packet and the matching source. A live LinkedIn call is optional. Do not check out a PR or attempt a fix. That is /3-verify-pr-fix.
NUM=$(echo "$ARGUMENTS" | sed -E 's|.*/||; s|#||g' | grep -oE '^[0-9]+' | head -1)
[ -z "$NUM" ] && { echo "Invalid input: '$ARGUMENTS'. Pass an issue number or URL." >&2; exit 1; }
REPO=stickerdaniel/linkedin-mcp-server
gh issue view $NUM --repo $REPO --comments
From the thread extract the tool, arguments, observed result, runtime, LinkedIn variant, and related issues. Distinguish observations, source findings, hypotheses, and work not run.
Map the tool to code:
linkedin_mcp_server/tools/<surface>.py. MCP entrypoint and arg validationdocs/linkedin-architecture.md. Generated ownership table; follow it to linkedin_mcp_server/linkedin/<owner>.py rather than treating linkedin/extractor.py as the implementationlinkedin_mcp_server/linkedin/fields.py. PERSON_SECTIONS / COMPANY_SECTIONS (each entry = one navigation)tests/linkedin/test_<owner>.py; use tests/test_fields.py, tests/test_identifiers.py, and tests/test_link_metadata.py for those owners, and tests/linkedin/test_facade_*.py only for facade contractsInspect the relevant source at the current SHA. Do not invent a LinkedIn result or an unimplemented tool call.
This step is complete when the report names the issue, the code SHA, the inspected evidence, the facts established, the unverified runtime claims, and the next decision.
Use one of:
A successful call on the maintainer's different account never refutes a failure on the reporter's account. Target content language is not the authenticated account's UI language. Captures and URL or attribute evidence may establish the needed variation without another live call.
Review reporter commands before execution. Publishing permission does not authorize send_message, a connection request, a forced login, or repeated calls with account side effects. Record sent, recipient_selected, and retry_safe as observed fields, not as replay authorization.
test cases passed, vs 5 / 14 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.