browser-search-and-handoff — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited browser-search-and-handoff (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Run a repeatable search workflow across websites where the user defines target attributes, Codex gathers matching entities, and the process ends in a controlled handoff before sensitive or anti-bot steps. Use this skill to separate what the AI can do reliably from what the human must confirm or complete manually.
Start by determining:
If the user has not already provided criteria, stop and ask for them before opening sites. Extract:
Use criteria-patterns.md for domain-specific intake patterns.
Default to a local CSV derived from tracker-template.csv unless the user explicitly asks for a different storage target. Prefer a local file first because it is easy to inspect, diff, and resume.
Use Google Sheets only when the user asks for it or when collaborative editing matters.
Track both discovery and follow-up state in the same record. Each row should represent one candidate entity plus its next-action page.
Use tracker-fields.md for the default schema.
For live browser work, prefer the agent-browser skill and its CLI workflows over ad hoc browsing instructions.
While searching:
Do not over-filter too early. Keep borderline matches if the user may want to review them later.
For every promising entity, store:
Prefer transparent reasoning over opaque scoring. A simple high, medium, low fit label is usually enough if paired with evidence.
After finding a matching listing, locate the page where the real action happens:
Store both the discovery URL and the action URL. They are often different and both matter for resuming work later.
Also capture blockers:
Treat this skill as human-in-the-loop by default.
AI-safe tasks usually include:
Human-only tasks usually include:
Handle credentials conservatively:
When automation must pause, return a compact handoff package:
If the user wants to resume later, continue from the tracker rather than rediscovering the same listings.
Use criteria-patterns.md to adapt the intake and scoring model to apartments, jobs, partner search, or another domain.
Use tracker-fields.md to decide which columns must be populated before handing off a shortlist.
For job applications, this skill owns discovery, scoring, shortlist tracking, and action URL collection. Once the user selects a specific job/application URL to execute, switch to job-application-operator for schema-driven field inventory, private-profile loading, document upload, logging, and submit-or-stop decisions.
Expect these problems and account for them explicitly:
When blocked, document the blocker in the tracker and move to the next candidate instead of losing momentum.
Prefer concise, structured outputs over long prose dumps.
For each result, capture at minimum:
Use the tracker as the source of truth. Summaries should be derived from it, not the other way around.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.