Repository navigation
Closed
Labels
approvedSubmission was approved by a maintainerSubmission was approved by a maintainerexternal-pluginPublic external plugin submissionPublic external plugin submissionneeds-review:MEDIUMContributor reputation check flagged MEDIUM riskContributor reputation check flagged MEDIUM risk
Description
Plugin name
wraith
Short description
Skills and agents for data and infrastructure engineering on GCP: dbt on BigQuery, ODCS data contracts, Cloud Run and Terraform. Reviews data and infrastructure plans before anyone builds them and writes schema-valid data contracts.
GitHub repository
wraith-proxy/wraith-skills
Plugin path inside the repository
plugin
Ref to review
v0.8.3
Commit SHA to review
d7e2c3caa203731e030ec63fe273638b9fd92c59
Version
0.8.3
License identifier
MIT
Author name
wraith-proxy
Author URL
https://github.com/wraith-proxy
Homepage URL
No response
Keywords
data-engineering
gcp
bigquery
dbt
cloud-run
terraform
odcs
data-contracts
ontology
agents
Additional notes for reviewers
This is a resubmission after #6233 (v0.5.7, closed 2026-11-09). Each reason given there, and what changed:
- Mostly workflow adapted from other plugins. Since v0.8.0 the generic
process skills are gone. wraith now ships only its data and GCP skills
and agents, and names obra/superpowers and mattpocock/skills as optional companions instead of carrying adapted copies (ADR 0006 in the repo). - Session-start hook and command guard. Both were removed in v0.6.0.
wraith has no hooks and no MCP servers; it is Markdown skills and
agents only. - No benchmark results. 4. No Copilot uplift shown. We ran the benchmark on GitHub Copilot CLI (wraith v0.8.3, Copilot Fusion), with and without wraith, 5 runs per case per arm:
- Plan review, planted problems found: 76.7% with wraith, 49.4%
without (180 per arm; higher in 11 of 12 cases). - ODCS data contract valid on first write: 64% with, 4% without.
- Destructive-command guard: no uplift on Copilot (0% stopped with,
4% without). We report it as a limit; Copilot's own tool approval is
the control to rely on there.
Blind audit of the judge (72 runs, 36 per host): the maintainer
agreed with it on 88.9%. All 8 disagreements were the judge being
stricter, never more lenient. The Copilotwithoutarm is probably
undercounted, so the Copilot plan-review gap is likely overstated.
On Claude Code the same plan review scores 90.0% with, 63.9% without.
Method, tallies and tally scripts (traces are not published):
https://github.com/wraith-proxy/wraith-benchmarks
Write-up: https://github.com/wraith-proxy/wraith-benchmarks/blob/main/docs/2026-11-09-two-hosts.md
- Plan review, planted problems found: 76.7% with wraith, 49.4%
The plugin now lives in plugin/ (ADR 0007), hence the plugin path.
Submission checklist
- The plugin lives in a public GitHub repository.
- The ref and/or sha I provided is immutable (release tag and/or full 40-character commit SHA), not a branch.
- This submission follows this repository's contribution, security, and responsible AI policies.
- This plugin is not already listed in the Awesome Copilot marketplace.
Reacted by Thomas Peoples
Metadata
Metadata
Assignees
Labels
approvedSubmission was approved by a maintainerSubmission was approved by a maintainerexternal-pluginPublic external plugin submissionPublic external plugin submissionneeds-review:MEDIUMContributor reputation check flagged MEDIUM riskContributor reputation check flagged MEDIUM risk
Activity
github-actions commentedon Nov 10, 2026
🟡 Contributor Reputation Check: MEDIUM risk
Maintainers: please review this contributor before merging.
See the workflow run for full details.
Automated check powered by AGT.
github-actions commentedon Nov 10, 2026
✅ External plugin intake passed
This submission passed automated intake validation and quality checks and is ready for maintainer review.
View workflow run · Download full quality gate logs
v0.8.3d7e2c3caa203731e030ec63fe273638b9fd92c59Reviewer signals
These are non-blocking heuristics for maintainer review; they are not evidence of misconduct or low quality.
Quality gate summary
Legend: ✅ pass ·⚠️ warning · 🛑 fail
spec compliance output (✅ pass)
vally lint output (✅ pass)
install smoke test output (✅ pass)
Version match output
Ref/SHA consistency output
Canvas structure output
Canonical external.json payload
{ "name": "wraith", "description": "Skills and agents for data and infrastructure engineering on GCP: dbt on BigQuery, ODCS data contracts, Cloud Run and Terraform. Reviews data and infrastructure plans before anyone builds them and writes schema-valid data contracts.", "version": "0.8.3", "author": { "name": "wraith-proxy", "url": "https://github.com/wraith-proxy" }, "repository": "https://github.com/wraith-proxy/wraith-skills", "license": "MIT", "keywords": [ "data-engineering", "gcp", "bigquery", "dbt", "cloud-run", "terraform", "odcs", "data-contracts", "ontology", "agents" ], "source": { "source": "github", "repo": "wraith-proxy/wraith-skills", "path": "plugin", "ref": "v0.8.3", "sha": "d7e2c3caa203731e030ec63fe273638b9fd92c59" } }wraith-proxy commentedon Nov 10, 2026
A real human being who took on the last feedback given and has now narrowed the scope to a niche data engineering plugin, without the bloat of Matt Pocock's skills, or Obra's superpowers. We now also have benchmarks in Copilot published. It looks decent, too
aaronpowell commentedon Nov 11, 2026
/approve
This addresses the earlier feedback: the generic workflow bundle and hooks are gone, and the focused data/GCP skills now have with/without Copilot results at this exact pin. The ODCS validation and data-review constraints show concrete uplift. Please update the root README to the new results and keep the limitations visible, especially the judge bias and the lack of Copilot guard uplift; native tool approvals remain the safety boundary.
github-actions commentedon Nov 11, 2026
✅ External plugin approved
A maintainer approved wraith, and the submission issue has been closed.
wraith-proxy/wraith-skillsmain: [external-plugin] Add wraith #6361