Skill Tracer
Skills › Knowledge Work Plugins › job-post-builder › gotchas.md

gotchas.md

68 lines · small-business/skills/job-post-builder/reference/gotchas.md at 2574e29

Open on GitHub

Most marks come from matching words in the file, so treat them as pointers.

1# Gotchas
2

Gotchas

Raw lines 1 to 2
3## 1. DocuSign login state
4
5✗ **Bad:** Claude navigates to `https://app.docusign.com` and immediately tries
6to upload the offer letter without checking whether the user is logged in.
7
8✓ **Good:** Immediately after navigating to DocuSign, check for a login screen.
9If a login prompt appears, pause and ask the user to log in, then wait for
10confirmation before continuing.
12**Why it matters:** DocuSign redirects unauthenticated sessions silently. If the
13login check is skipped, the automation tries to click UI elements that don't exist
14and fails mid-flow with no clear error.
16---

1. DocuSign login state

✗ Bad: Claude navigates to https://app.docusign.com and immediately tries to upload the offer letter without checking whether the user is logged in.

✓ Good: Immediately after navigating to DocuSign, check for a login screen. If a login prompt appears, pause and ask the user to log in, then wait for confirmation before continuing.

Why it matters: DocuSign redirects unauthenticated sessions silently. If the login check is skipped, the automation tries to click UI elements that don't exist and fails mid-flow with no clear error.

Raw lines 3 to 17
Involves a person pause and ask :9
18## 2. Missing candidate details before entering the browser
20✗ **Bad:** Claude reaches Phase 6, opens DocuSign in Chrome, and then asks mid-flow:
21"What's the candidate's email address?"
23✓ **Good:** If the user chose DocuSign delivery in Phase 1, collect the candidate's
24full name and email address before Phase 6 begins — either in Phase 1 or at the
25end of Phase 5. Never enter the browser flow without both fields.
27**Why it matters:** Interrupting an open browser session to collect missing data
28disrupts the automation state and confuses the user.
30---

2. Missing candidate details before entering the browser

✗ Bad: Claude reaches Phase 6, opens DocuSign in Chrome, and then asks mid-flow: "What's the candidate's email address?"

✓ Good: If the user chose DocuSign delivery in Phase 1, collect the candidate's full name and email address before Phase 6 begins — either in Phase 1 or at the end of Phase 5. Never enter the browser flow without both fields.

Why it matters: Interrupting an open browser session to collect missing data disrupts the automation state and confuses the user.

Raw lines 18 to 31
32## 3. Re-asking for context the user already provided
34✗ **Bad:** The user says "we need to hire a senior PM, fully remote, USD 160–180k"
35and Phase 1 asks for role title, location, and compensation anyway.
37✓ **Good:** Extract role title, location, and compensation from the message, confirm
38them in a single sentence, and ask only for the fields that are genuinely missing.
40**Why it matters:** The skill explicitly requires "one focused clarifying question
41rather than a long form." Redundant questions break trust and slow the workflow.
43---

3. Re-asking for context the user already provided

✗ Bad: The user says "we need to hire a senior PM, fully remote, USD 160–180k" and Phase 1 asks for role title, location, and compensation anyway.

✓ Good: Extract role title, location, and compensation from the message, confirm them in a single sentence, and ask only for the fields that are genuinely missing.

Why it matters: The skill explicitly requires "one focused clarifying question rather than a long form." Redundant questions break trust and slow the workflow.

Raw lines 32 to 44
45## 4. Silently expanding the user's existing format
47✗ **Bad:** The user has a 3-section job post on file. Claude produces a 7-section
48post based on `job-post-structure.md` without asking.
50✓ **Good:** Map the user's existing format against the reference, identify missing
51sections, and ask one question: "Your existing JD has X and Y — want me to add Z,
52or keep your current format?"
54**Why it matters:** The user's format is the source of truth. Overriding it silently
55may conflict with internal HR or legal standards the user hasn't mentioned.
57---

4. Silently expanding the user's existing format

✗ Bad: The user has a 3-section job post on file. Claude produces a 7-section post based on job-post-structure.md without asking.

✓ Good: Map the user's existing format against the reference, identify missing sections, and ask one question: "Your existing JD has X and Y — want me to add Z, or keep your current format?"

Why it matters: The user's format is the source of truth. Overriding it silently may conflict with internal HR or legal standards the user hasn't mentioned.

Raw lines 45 to 58
59## 5. Inventing compensation figures
61✗ **Bad:** No salary range was provided, so Claude writes "USD 120,000–USD 150,000 DOE"
62in the job post or offer letter.
64✓ **Good:** If compensation isn't provided, omit the range from the job post entirely.
65In the offer letter, use `[ANNUAL SALARY — confirm with HR]` as a bracketed placeholder.
67**Why it matters:** Inventing compensation figures creates legal and HR liability.
68The skill's instructions are explicit: "Don't invent a range."

5. Inventing compensation figures

✗ Bad: No salary range was provided, so Claude writes "USD 120,000–USD 150,000 DOE" in the job post or offer letter.

✓ Good: If compensation isn't provided, omit the range from the job post entirely. In the offer letter, use [ANNUAL SALARY — confirm with HR] as a bracketed placeholder.

Why it matters: Inventing compensation figures creates legal and HR liability. The skill's instructions are explicit: "Don't invent a range."

Raw lines 59 to 68