v1.5.0 update

v1.6

v1.4.0 update

V1.4.0 update

v1.3.0 update

v1.3.0 update

v1.2.0 release

update v1.2.0

Adding Skill Illustration Fig

v1.1.1

Add files via upload

Field Onboarding v1.1.0

v1.1.0 — Trigger scope and reference discipline

This release fixes two behavioral gaps in v1.0.0 and restores the imperative
register of the instruction text.

Negative triggers

v1.0.0 had broad positive triggers and no negative ones, so the skill over-fired:
a narrow factual question could be turned into a prerequisite checklist plus a
five-rung ladder.

SKILL.md now opens with When not to use this skill, covering narrow factual
questions, specialists asking inside their own field, explicit requests for a
short answer, non-comprehension tasks (translation, editing, search, debugging),
users blocked mid-experiment, and users who already declined the ladder.

The governing heuristic is the one-turn test: if a single turn answers the
question well, answer it and offer the ladder once, in one line. Answer first,
offer second.

The frontmatter description now carries a matching Do not use for ... clause
alongside the positive triggers.

Reference discipline

Reading paths are where models fabricate, and a plausible title with a plausible
year costs the reader an afternoon.

New Naming literature section. Every named paper, review, book, or package
falls into exactly one of two buckets and must be labelled:

• Verified — looked up in this session, with DOI, arXiv ID, or journal,
  volume, and page.
• From memory, unverified — believed to exist, not checked, and labelled in
  those words.

Bare unlabelled citations are not permitted. Identifiers that were not actually
retrieved are never attached: a fabricated DOI is worse than no DOI, because it
looks checked. Without search capability the skill states the limitation once,
marks everything unverified, and gives executable search pointers (venue, group,
query) instead of citations. Three verified items beat seven unverified ones.

Instruction register

Normative prose was replaced with direct rules throughout. guessing the user's level when a short calibration would materially improve the explanation is now
Skipping Step 0 and guessing at their level. Agents follow the short imperative
form more reliably. Several concrete details dropped in the v1.0.0 rewrite are
restored: the stated black box for oversized prerequisites, the false-friend
warning in Rung 2, and the anti-hedging rule.

Session behavior

Two common real-session cases are now handled explicitly:

• The user ignores the prerequisite checklist and says "go ahead". Do not re-ask.
  Assume learned it, state the assumption in one line, correct downward the
  first time an anchor fails to land.
• The user skips the checkpoint and says "continue". Do not re-ask. Advance, fold
  the diagnostic into the next rung's opening, lower the assumed level a notch.

Docs

• README.md: new Scope and Reference discipline sections, applicability
  step added to the core workflow, corrected repository tree, changelog.
• AGENTS.md: applicability check and the literature rule added to the agent
  entrypoint; capability mapping updated for agents without search.
• references/examples.md: new Example 1 (a prompt the skill should decline to
  take over) and Example 8 (a labelled reading path, with and without search).
• CONTRIBUTING.md: widening the trigger surface now requires a matching
  negative example; register guidance added.

Upgrading

Drop-in replacement. Re-package field-onboarding/ as your skill root, or pull
main. No config changes.

Full diff: https://github.com/ljx-chase/research-field-onboarding/compare/v1.1.0

Field Onboarding v1.0.1

Update README for public release

Added badges for license, version, and compatibility to README. Updated quick start instructions for ChatGPT and Codex agents. Removed GitHub publishing instructions.