skills.rest
stellar-experimental
stellar-experimentalOFFICIAL
Explore all skills by this author →
GitHub

truth-maintenance

Coordinates multi-lane maintenance rounds across drift, evals, goldens, and upstream issue tracking.

8|6|Updated Jul 31, 2026
Search Tags:#code review#github#drift detection#maintenance#orchestration#ci#evals
One-click install
npx skills add https://github.com/stellar-experimental/stellar-raven --skill truth-maintenance-stellar-experimental
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: truth-maintenance
Source: https://github.com/stellar-experimental/stellar-raven/tree/main/.agents/skills/truth-maintenance
Command: npx skills add https://github.com/stellar-experimental/stellar-raven --skill truth-maintenance-stellar-experimental

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Keeping a repository's documented truth current requires checking live service drift, eval results, golden answers, and upstream issues simultaneously, which is too much for a single review pass. This Skill orchestrates that work as a coordinated maintenance round with parallel lanes and a shared ledger. ## Core Features & Use Cases - Multi-lane orchestration: Spawns separate reviewer agents for drift, eval, golden-truth, and improvements lanes, each appending verdicts to a dated ledger at .agents/rounds/. - Stale-gospel queue management: Triages past-due truth.reverifyBy dates flagged by npm run eval:qa:lint -- --stale, re-verifying or recording dated extensions with root causes. - Upstream issue/PR follow-up: Tracks GitHub issues and PRs the repo opened, re-runs original triggers live, and records deterministic status tables with next wake-up dates. - Use Case: When a daily CI job fails because golden answers went stale, use this Skill to dispatch re-verification across lanes, reconcile cross-lane effects, and close out with a complete ledger of commands, verdicts, and follow-ups. ## Quick Start Run a full truth-maintenance round to verify repo truth is current across drift, evals, goldens, and upstream PRs, and record the results in a dated ledger.

Frequently Asked Questions about truth-maintenance

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a full truth-maintenance round across multiple review lanes?▼

Open a dated ledger at .agents/rounds/, write a lane plan, then spawn one reviewer agent per lane (drift, eval, golden, improvements) using the herdr skill. Each agent appends verdicts and evidence to the ledger while the coordinator owns repo edits and final gates.

How do I triage stale golden answers flagged by CI?▼

Run npm run eval:qa:lint -- --stale to list past-due cases, then dispatch each through the golden-truth lane to re-verify with updated truth.verified and asOf dates, or record an explicit dated extension with a rootCause. Never silently bump dates.

When should I use truth-maintenance instead of a single lane runbook?▼

Use it when work crosses lanes, such as drift that changes facts consumed by goldens, or when asked to confirm overall repo truth. Single issues like one drift item or one eval run belong to the dedicated lane runbooks.

How are upstream GitHub issues and PRs tracked during maintenance?▼

Each finding gets a deterministic status table row with trigger, upstream ref state, PR checks, live re-check result, and next wake-up date. A finding is only marked fixed-upstream when a live re-run of the original trigger passes, not when an issue closes.

What are the limitations of delegating patches to spawned agents?▼

By default spawned agents only research and append to the ledger while the coordinator owns repo edits. Delegate patches only when the write set is narrow, explicit, and disjoint, and reviewers must independently re-derive findings rather than accept summaries.

Related Skills

Explore More Software Engineering Skills→
agenticgogolagenticgogol
1

baseline-runner

Generates promptfoo configs and runs evals to produce scored baseline results.

Community
Intermediate
baekenoughbaekenough
13

post-release-followup

Collect and categorize release findings into actionable follow-up recommendations.

Community
Advanced
sheawinklersheawinkler
78

requesting-code-review

Automate pre-commit verification with static scans, tests, linting, and reviewer gates.

Community
Advanced
sandecosandeco
1.5k

reversa-debugger-debate

Runs multi-agent debates with isolated judge to decide bug diagnosis, repair, or spec verdicts.

Community
Advanced