jd-decoder
Skill3 starsChanged 4 months agoMITView on GitHub
Use when the user shares a job posting and asks whether to apply, what the posting really means, whether a job is real or a ghost listing, what a salary band signals, or how to prioritize between several postings.
In this repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- cold-outreach-writerSkill
- company-researcherSkill
- mock-interviewerSkill
- resume-tailorSkill
- salary-negotiatorSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
No named team, manager, or concrete problem.
- Distinctive: line 17
Not listed on the company's own careers page (job-board only)
- Distinctive: line 19
No recruiter activity visible (no LinkedIn posts about the role, no outreach happening)
- Distinctive: line 51
Must-haves vs wishlist (two short lists)
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
JD Decoder
Treat every posting as a probabilistic signal, not a promise. Some listings are stale, evergreen, exploratory, duplicated, or no longer funded, but published estimates are not interchangeable and should not be turned into one universal “ghost job” rate. The practical takeaway is to inspect the specific posting and company, then scale effort to the evidence.
Process
Step 1: Score ghost risk before anything else
One point each (0-6):
- Posted or reposted for 45+ days with no edits
- No named team, manager, or concrete problem; responsibilities read as template
- Not listed on the company's own careers page (job-board only)
- Company had layoffs in the last 6 months in this function
- No recruiter activity visible (no LinkedIn posts about the role, no outreach happening)
- Extremely wide or missing comp band with no explanation
0-1: real, tailor deeply. 2-3: probably real, tailor moderately and verify. 4+: verify before investing; offer to draft a short note to the likely hiring manager asking "is this role actively interviewing?" That question is legitimate, cheap, and converts uncertainty into information. A repost alone is NOT proof of fake: budgets change, finalists fall through, some roles are evergreen.
Step 2: Classify every requirement
Four piles: must-have (early, repeated, connected to responsibilities), signal (tells you the team's stack and pain), nice-to-have (long comma lists; candidates matching 100% of a wishlist do not exist and the hiring manager knows it), compliance filler. Tailor only to must-haves and signals.
Step 3: Translate loaded language into diligence questions, not verdicts
Posting-language studies are weak evidence of anything, so convert phrases into interview questions instead of conclusions:
- "Fast-paced" → "What changed in the last two quarters that created this role?"
- "Wear many hats" → "Which three functions does this role own in the first 90 days?"
- "0 to 1 / founding" → equity must be a real conversation; if the comp section is silent on it, that is a question.
- A hyper-specific requirement ("migrating off LangChain") → they have that exact problem now; direct-hit evidence here is the user's opening line.
- One role asking for research + infra + customer-facing + data + sales engineering → unicorn trap; the role is underscoped in their heads. Generate clarifying questions and lower the tailoring investment.
Step 4: Decode the comp band
A very wide compensation band can hide multiple levels or geographies. The expert move is decoding, not avoiding: does it span multiple levels? Multiple geos? Is the top reachable for external hires? Flag which question applies and queue it for the screen. Band position also signals leveling; map level by scope (ownership, ambiguity, cross-functional impact), never by years alone. Years-of-experience lines are soft; direct production evidence can outweigh longer adjacent experience, and you should say when the user should apply anyway.
Step 5: Verdict
Exactly one of:
- Apply now: strong match on must-haves, low ghost risk. List the 2-3 evidence points to lead with.
- Stretch, apply anyway: 60-80% of must-haves and the gaps are learnable. Name each gap and its one-line honest framing.
- Verify first: good match but ghost risk 4+. Provide the verification note.
- Skip: fundamental mismatch or the decoded reality contradicts what the user wants. One sentence why, so the next "apply" verdict is trusted.
Be willing to say skip; a decoder that always says apply is useless. But remember the base rates: most postings are real, and false negatives are expensive in a scarce market.
Output format
- Ghost-risk score with the checklist shown
- What they actually need (3-5 bullets, plain language)
- Must-haves vs wishlist (two short lists)
- Decoded phrases and the diligence questions they generate
- Comp band read + leveling estimate, with what to verify
- Verdict with reasoning
Label every inference as an inference. If the user shares multiple postings, rank them by (fit × ghost-risk-adjusted odds) and say where to spend this week's effort.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

