preetamnath/agent-skills

vet-fact

Judge whether a candidate current-guidance fact earns a durable instruction line: keep only what prevents a likely wrong answer; cut derivable, historical, setup, or default restatements.

First seen Jun 13, 2026

Installation

$ npx skills add preetamnath/agent-skills --skill vet-fact

Summary

  • Judge whether a candidate current-guidance fact earns a durable instruction line: keep only what prevents a likely wrong answer; cut derivable, historical, setup, or default restatements.
  • TRIGGER when: user asks 'does this belong in AGENTS.md / a rule / the docs', 'is this worth keeping/writing down', or 'should this be a comment'; vetting or pruning current guidance before placement or shaping.
  • SKIP when: evaluating or placing active state or dated evidence (place-fact).

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from preetamnath/agent-skills · top by installs.

npx skills add preetamnath/agent-skills

Browse all from preetamnath/agent-skills

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

Stars 4
License LICENSE.md
Default branch main
Open issues 0
Status Active

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 2,637 B
  • docs SUMMARY.md 492 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 5 installs

SKILL.md

Vet Fact

Primitive: WORTH — does this candidate fact deserve a current-guidance line at all?

Steps

  1. State the fact in one line, then apply the gate: would a capable future agent likely get the wrong answer without it? Cut it when code, configuration, or live help reveals it reliably; lookup cost alone does not justify current guidance.
  2. Keep WORTH inside current guidance:

- Exclude state and evidence records. Do not run an active-state document or dated investigation through this lens; apply WORTH only to conclusions proposed for current guidance. - Preserve freshness provenance. Keep a per-entry Discovered: YYYY-MM (sha) stamp only when it lets a future agent judge whether a current workaround's premise is stale; a trim must preserve it.

  1. Cut a current-guidance candidate on any hit:

- Setup. Put setup and onboarding in the README. - History. Cut dates, plan IDs, commit SHAs, superseded/originally notes, completion summaries, and deferred-item lists. - Defaults. Cut restated framework or harness defaults. - Drift-prone literals. Cut hex codes, version pins, and exact syntax unless the literal is itself load-bearing.

  1. Keep only a recognized category:
Category Keep when
Gotcha A non-obvious trap, anti-pattern, or dead-code warning can cause a wrong change.
Coupling A cross-file constraint is invisible from any single file.
Convention The required choice contradicts the obvious default.
Rationale A tempting wrong alternative exists; keep the reason as the constraint.
Pointer The agent must follow the target before touching the governed work.
  1. Cut facts already carried elsewhere. WORTH is corpus-relative: an existing comment, rule, or document makes the same fact a duplicate. Catch duplicates visible in scope; flag cross-document deduplication when the needed sources are outside scope.
  2. Return the verdict: keep or cut, with a one-line reason; on a keep, include the category (gotcha / coupling / convention / rationale / pointer).