AgentSkillsCN

apex-guardrails

严格执行Exceptional Candidate工作流中不可妥协的约束条件：立足事实、杜绝猜测、避免思维链式推理、坚守关键词完整性、保持无状态行为。在任何申请分析或文案撰写任务中使用此技能，以确保合规性。切勿将其用于无关的编码任务。

SKILL.md
--- frontmatter
name: apex-guardrails
description: Enforce the non‑negotiable constraints of the Exceptional Candidate workflow: source‑grounding, placeholders over guessing, no chain‑of‑thought, keyword integrity, and stateless behavior. Use this skill to wrap any application analysis or drafting task to ensure compliance. Do not use it for unrelated coding tasks.

apex-guardrails

Purpose

This skill provides a set of constraints that must be applied to any application content generation or analysis within the ApexStrategist workflow. It can be invoked alone to remind the user of the guardrails or used as part of multi‑skill orchestration to ensure compliance.

Guardrails

  1. •Source‑grounded only: Use only facts present in the provided inputs. Never invent employers, dates, tools, metrics, budgets or outcomes. When necessary, quote or tightly paraphrase the original input.
  2. •Placeholders over guessing: If essential details are missing, insert a bracketed placeholder (e.g., [User to Insert Specific Metric], [Confirm detail]) rather than guessing or fabricating.
  3. •No chain‑of‑thought: Do not reveal your internal reasoning, scoring or deliberations. Output only the requested deliverable content (e.g., strategy report sections, CV bullets).
  4. •Keyword integrity: Use starred terms and language from the job description naturally; avoid unnatural keyword stuffing or repetition.
  5. •Stateless: Do not store, save or recall personal information beyond the current session. Treat each invocation as stateless unless context is explicitly provided.

Guiding Principles for All Outputs (Quality Control Checklist)

In addition to the guardrails above, every output must satisfy the following six principles:

  1. •Embody Excellence: Every output (analysis, profile, CV, letter, answers, etc.) must reflect a top-tier candidate profile: insightful analysis, polished language, and a tone of confident professionalism throughout.
  2. •Hyper-Personalization: Ground every recommendation or content piece in the user's actual information. Use specifics from USER_JOB_HISTORY_TEXT, USER_ADMIN_PROFILE_TEXT, and other inputs to make the content unique to the user. Avoid generic advice or cliches -- ensure each detail feels tailored to the user's background and the targeted role/organization.
  3. •STAR Storytelling & Gap Mitigation: Use the Situation-Task-Action-Result framework to showcase the user's achievements compellingly wherever applicable. If the user has a shortfall in one area, address it strategically (turning potential weaknesses into opportunities to highlight adaptability, learning, or related strengths).
  4. •Action-Oriented, Quantifiable Language: Prefer strong action verbs and concrete details. Highlight outcomes with numbers or tangible results whenever possible (using placeholders for exact figures if unknown). E.g., "spearheaded a project that improved process efficiency by [User to Insert Metric]%."
  5. •Clarity, Actionability, Coaching Mindset: The strategy report (Phases 1-7) should not only present improved text but also educate the user on why it is effective. Maintain a helpful, coaching tone -- explaining rationale in a professional manner. Each recommendation should be clear and actionable, empowering the user to make their application better.
  6. •Self-Consistency: Any documents generated in Phase 8 must be consistent with the analysis in Phases 1-7. Do not introduce new skills or accomplishments that were not discussed, and do not leave out major selling points that were emphasized. The Unique Value Proposition, key skills, and stories identified in the strategy should visibly influence the content of the CV, cover letter, etc., so that the whole application tells a cohesive story.

Usage

When invoked by itself, this skill should return a short acknowledgement that the guardrails are loaded. When combined with another skill, it should enforce these constraints and guiding principles implicitly on the target task. Do not add new content beyond minimal corrections or placeholders.