diff --git a/.agents/skills/gh-pr-results-review/SKILL.md b/.agents/skills/gh-pr-results-review/SKILL.md
new file mode 100644
index 00000000..553772d2
--- /dev/null
+++ b/.agents/skills/gh-pr-results-review/SKILL.md
@@ -0,0 +1,71 @@
+---
+name: gh-pr-results-review
+description: Review GitHub pull request CI outcomes with the GitHub CLI and extract actionable failure details from GitHub Actions logs. Use when asked to check failing PR checks, inspect lint/test/type-check errors, identify failing jobs, or summarize why a PR pipeline failed and what to fix first.
+---
+
+# GH PR Results Review
+
+## Workflow
+
+1. Identify the PR number and check status.
+- Run:
+```bash
+gh pr checks <pr-number> 2>&1
+```
+- Capture failing checks, durations, run URL, and job URLs.
+- If no failures exist, report success and stop.
+
+2. Resolve run and failing job IDs.
+- Derive the workflow run ID from the run URL when available.
+- Derive each failing job ID from job URLs, or list jobs with:
+```bash
+gh run view <run-id> --json jobs
+```
+
+3. Pull logs for failed jobs.
+- For each failed job:
+```bash
+gh run view <run-id> --job <job-id> --log 2>&1 | tail -80
+```
+- If the tail misses the error, inspect more lines or grep for error markers:
+```bash
+gh run view <run-id> --job <job-id> --log 2>&1 | rg -n "error|failed|traceback|E999|F401|mypy|pytest"
+```
+
+4. Summarize root causes with evidence.
+- Report by failing check/job name.
+- Include the shortest exact evidence line(s) that explain failure.
+- State likely fix direction (for example: ruff import ordering, mypy type mismatch, failing pytest assertion).
+
+## Output Format
+
+Use this structure:
+
+```text
+check the CI failure details.
+
+Reviewed GitHub Actions logs for test and linting results
+Bash
+gh pr checks <pr-number> 2>&1
+<key failing checks output>
+
+Bash
+gh run view <run-id> --job <job-id> --log 2>&1 | tail -60
+<relevant error/failure excerpt>
+
+Bash
+gh run view <run-id> --job <job-id> --log 2>&1 | tail -40
+<relevant error/failure excerpt>
+```
+
+Then provide a short diagnosis section:
+- `Finding`: one sentence per failed job.
+- `Evidence`: command + key line.
+- `Next fix`: concrete first edit to attempt.
+
+## Guardrails
+
+- Do not claim a job failed without a log line proving it.
+- Prefer job-specific logs over workflow-level summaries.
+- Keep excerpts tight; include only lines needed to justify the finding.
+- If GitHub CLI lacks permission or run is missing, state the blocker and requested access explicitly.
diff --git a/.agents/skills/gh-pr-results-review/agents/openai.yaml b/.agents/skills/gh-pr-results-review/agents/openai.yaml
new file mode 100644
index 00000000..c2509d52
--- /dev/null
+++ b/.agents/skills/gh-pr-results-review/agents/openai.yaml
@@ -0,0 +1,4 @@
+interface:
+  display_name: "PR Results Reviewer"
+  short_description: "Inspect failing PR checks and logs"
+  default_prompt: "Use $gh-pr-results-review to inspect GitHub PR check failures, read failing job logs, and summarize root causes with exact evidence."