| name | watchouts-sweep |
| description | Cross-check a drafted application document (cover letter, statement of claims, targeted pitch, recruiter response, screening-question response) against the candidate's watch-outs register (Examples_Section_7_Watchouts.md) before delivery. Use whenever a draft is about to be saved to Submissions/ or presented as the final output of an application-drafting session. Also use when the candidate asks to 'sweep this for watchouts', 'check this for overclaim', or 'verify this against Section 7'. The sweep is mechanical (~30–60 sec) and catches the most common executive-application failure mode: an overclaim that the register would have caught. |
Watchouts Sweep — Section 7 verification skill
Setup note: This is a template skeleton. The sweep runs against the watch-outs register your framework accumulates in Examples_Section_7_Watchouts.md. The register starts thin — every entry is added after a real overclaim is caught (that is the discipline that makes it valuable). The skill needs no per-candidate editing beyond placeholders.
When to invoke
- Default trigger: before saving any drafted
.docx to [Workspace Folder Path]/Submissions/, before delivering any screening-question response, and before any final application document is presented to the candidate.
- Explicit triggers: "sweep this for watchouts", "check for overclaim", "verify against Section 7".
- Skip when: the draft is an intermediate iteration under active review (sweep at the end, not every revision).
Inputs
- The draft text (inline or file path).
- The matched matrix row ID — its
Notes / Distinct from column names the specific 7.1.x sub-variants that apply. If unknown, infer the role family and sweep against family-wide watch-outs plus all named sub-variants.
- The role context (organisation, sector, role title) — for domain-specific watch-outs.
Method
- Load the register — read
Examples_Section_7_Watchouts.md in full: 7.1 experience-limits table and sub-variants, 7.2 framing discipline, 7.3 evidence-recency flags.
- Filter to applicable watch-outs using the matched row and role context. Universal honesty boundaries in the 7.1 table always apply.
- Sweep the draft — for each applicable watch-out, search the draft for breaching claims: scope inflation, temporal-qualifier loss ("first" without "at the time"), attribution drift (organisation-level outcomes claimed as personal), stale metrics (7.3).
- Score each finding High / Medium / Low with exact-quote evidence and a register-compliant alternative wording.
Output and gate behaviour
Findings list (severity / exact quote / breached watch-out / recommended rewording), then verdict. High-severity findings block delivery until amended; Medium and Low are advisory — the candidate decides. (Run advisory-only for the first two weeks of a new setup while the register is young, then flip High to blocking.)
