work-recommendations

Resolves ordered survey-work plate items to GitHub issues and dispatches them through take-it.

Updated Mar 26, 2026
One-click install
npx skills add https://github.com/Sassy-Dog/sassydog-skills --skill work-recommendations-sassy-dog
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: work-recommendations
Source: https://github.com/Sassy-Dog/sassydog-skills/tree/main/skills/work-recommendations
Command: npx skills add https://github.com/Sassy-Dog/sassydog-skills --skill work-recommendations-sassy-dog

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a prioritized work plate into shipped pull requests normally means manually filing issues for unfiled items, checking labels and security constraints, and dispatching work in the right order. This Skill automates that translation: every plate item gets exactly one disposition (DISPATCH, FILE, SHEPHERD, HOLD, CONFIRM-EACH, or HUMAN-ONLY) and the dispatchable ones go to take-it in plate order behind a single approval. ## Core Features & Use Cases - Handle-based disposition: Resolves each plate item's handle (#N, pr:#N, sentry:<id>, fire-watch:<kind>/<slug>, or none) to a disposition, with live label checks that route security-labeled issues to CONFIRM-EACH and auto-security-watch findings to HUMAN-ONLY. - Idempotent issue filing: Files issue-less items through file-or-link-issue.sh with dedupe markers (sentry-source:, plate-source:, fire-watch-source:) so re-runs never create duplicates. - Single preview and approval: Shows one dry-run preview covering all filings, take-it dispatches, and pr-shepherd merges before any outward write happens. - Use Case: After running survey-work, say "work the recommendations" to file the two issue-less tech-debt items, shepherd a green Dependabot PR, and dispatch the top five issues to parallel take-it sub-agents in plate order. ## Quick Start Ask the agent to work the recommendations from the latest survey-work plate in this repository.

Frequently Asked Questions about work-recommendations

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

FAQPage Schema
How do I work the recommendations from a survey-work plate?

Invoke the skill with a phrase like "work the recommendations" after survey-work renders its plate. It resolves each item's handle, files issue-less items behind one preview, and dispatches up to five issues to take-it in plate order after your approval.

How does the skill decide which plate items to dispatch versus hold?

Each item's handle and a live label read decide: plain issues are DISPATCH, `security`-labeled issues are CONFIRM-EACH, `auto-security-watch` findings are HUMAN-ONLY, off-site issues are HOLD, and PRs passing the held-PR check are SHEPHERD. A failed label read is treated as UNKNOWN and held.

Does it work without the take-it config file?

No. The skill reads the repo's .claude/sassy-dog/take-it.md and stops on NO_CONFIG before filing anything, directing you to run sassy-dog:setup-config first. This prevents filing issues that take-it would then refuse to dispatch.

Can it create duplicate GitHub issues on re-runs?

No. All filing goes through file-or-link-issue.sh with dedupe markers (`sentry-source:`, `plate-source:`, `fire-watch-source:`), so a re-run links the existing issue instead of creating a new one. Direct `gh issue create` is explicitly forbidden.

What happens to security findings and secret-scanning alerts?

Secret rotation and revocation items are HUMAN-ONLY: surfaced first with their Fix line and never filed or dispatched, since an issue describing an active leak is itself a disclosure surface. Other `security`-labeled issues require explicit per-item confirmation in the approval.

What are the limitations on batch size per run?

At most five DISPATCH handles go to take-it per run; items six onward are listed under "Not dispatched this run" with the exact `take #...` command to continue. More than five would-file items triggers a stop to ask whether an umbrella issue fits.