From agent-harness
Use when asked to report harness feedback, promote captured guardrail failures into issues, or check "anything worth turning into a rule?". Clusters unreported feedback events and raises digest issues on the agent-harness repo.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-harness:harness-reportThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Promote captured feedback events into digest GitHub issues on the harness
Promote captured feedback events into digest GitHub issues on the harness repo — one issue per recurring failure pattern, so each can become a new guardrail. Capture is local and automatic; reporting is batched, reviewed by the user, and explicit.
Fold the append-only store to current-state-per-id with the helper (it
defaults a missing kind to gate-failure for back-compat):
& "${CLAUDE_PLUGIN_ROOT}/scripts/fold-events.ps1" | ConvertFrom-Json
Skip events that already have reportedIssue. If nothing is unreported,
say so and stop.
Group unreported events by recurring mistake, clustering within a single
kind — never mix gate-failure and review-comment in one cluster.
gate-failure: same gate + template + similar failure text/note. Read
the diffs/<id>.failure.patch / .fix.patch sidecars when the
outputTail is not enough.review-comment: same template + similar guidance across PRs/files. Read
the diffs/<id>.hunk.patch sidecar for the code the comment was on.
These are the highest-value clusters — a comment a human reviewer had to
leave is a rule the toolchain did not enforce.For each review-comment cluster, propose a remedy type from the
comment bodies and file paths:
tests/*.Architecture) or a dependency-cruiser rule (expo);Singletons are normally not worth reporting — mention them and let the user decide.
For each cluster, draft an issue body containing ONLY: occurrence count,
gate, template(s), project directory NAMES (not full paths), event ids,
the one-line notes, and a one-paragraph hypothesis of the rule that would
have prevented it. NO raw diffs, NO output tails, NO code — the harness
repo is public and events may come from private projects; diffs stay on
disk, referenced by event id. For review-comment clusters, also include
the proposed remedy type and the reviewer guidance summarized in your own
words — never paste raw diff hunks (sidecars stay local; the harness repo
is public). Show the user every title+body and get approval before posting
anything.
For each approved cluster:
gh issue create --repo ryan75195/dotnet-agent-harness --title "<short pattern summary>" --body "<digest>"
Then mark every event in the cluster:
& "${CLAUDE_PLUGIN_ROOT}/scripts/mark-reported.ps1" -EventId <id> -IssueUrl <url>
Report: issues raised (links), clusters skipped and why, events remaining unreported.
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.