From agent-harness
Use after a PR merges, or on request, to capture human review comments from a PR into the harness feedback store ("capture the review comments", "harvest PR feedback"). Routed from the merge step.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-harness:harness-capture-reviewThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Capture **human** PR review comments into the feedback store as
Capture human PR review comments into the feedback store as
review-comment events, so harness-report can later turn recurring
review feedback into new guardrails. Bots are skipped; capture is
idempotent (deduped on the GitHub comment id) and never blocks the merge
that preceded it.
Determine the PR. Use the explicit PR number if given, else the PR for the current branch:
gh pr view --json number,url,author
Resolve owner/repo with gh repo view --json nameWithOwner.
Fetch human review data (two sources):
gh api "repos/<owner>/<repo>/pulls/<pr>/comments" --paginate
gh api "repos/<owner>/<repo>/pulls/<pr>/reviews" --paginate
The first is inline comments (fields id, user.login, user.type,
path, line, body, diff_hunk, html_url). The second is review
summaries (use entries with a non-empty body; user, state,
html_url).
Normalize both into one array where each element has: commentId
(the numeric id), authorType (user.type), author (user.login),
isPrAuthor (true when user.login equals the PR author), body,
file (path or null for summaries), line (or null), diffHunk
(diff_hunk or null), commentUrl (html_url). Write it to a temp
JSON file.
Capture:
& "${CLAUDE_PLUGIN_ROOT}/scripts/capture-review.ps1" -InputJson <temp> -Pr <pr> -PrUrl <url>
Report the one-line summary the script prints. If any gh call
fails (no PR, no auth, offline), say so and stop — do not retry in a
loop and do not treat it as fatal to the surrounding work.
npx claudepluginhub ryan75195/dotnet-agent-harness --plugin agent-harnessSearches MemPalace before answering questions about past work, people, projects, or prior decisions. Returns verbatim stored content instead of guessing from model memory.
Guides Payload CMS config (payload.config.ts), collections, fields, hooks, access control, APIs. Debugs validation errors, security, relationships, queries, transactions, hook behavior.
Implements vector databases with Pinecone, Weaviate, Qdrant, Milvus, pgvector for semantic search, RAG, recommendations, and similarity systems. Optimizes embeddings, indexing, and hybrid search.