evaluate-role
Skill2 starsChanged 3 months agoMITView on GitHub
Work through one job in depth — what it actually is, fit against the evidence bank, the interview loop, the trade — and record the decision. Use when a req or JD is pasted, when asked "should I apply / is this worth it / explain this job", or to go deeper on a pick from a triage run.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- kevinmmiddleton/recruiter-bfftheir own version, first seen Oct 17, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- kevinmmiddleton/recruiter-bff★ 3same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- apply-formSkill
- cover-letterSkill
- job-triageSkill
- publishSkill
- research-companySkill
- setupSkill
- sync-appliedSkill
- tailor-cv-batchSkill
- tailor-cvSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 39
Read it to the end — comp, location, eligibility and the seniority bar are usually last.
- Very specific: line 41
If the source is an aggregator link, find the employer's own posting.
- Very specific: line 53
Run /research-company (engine: python -m research "<company>"), then close its open questions with your own web search and write them back …
- Very specific: line 81
Write docs/knowledge-base/personal/roles/<date>_<company>_<role-slug>.md.
- Distinctive: line 87
Update it in place as a thread continues.
- Very specific: line 89
When the decision is final, add the id to profile/skiplist.md with a one-line reason — that is the machine index, and it is what stops the …
- Distinctive: line 94
If the session produced a durable preference — a screening format that is a dealbreaker, a settled question, a gap that should always be …
- Very specific: line 108
Not a CV: When the decision is to apply, hand off to /tailor-cv (one) or /tailor-cv-batch (several) and /cover-letter.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Evaluate one role, in depth, and record it
Triage ranks the firehose. This is the other half: one role, in conversation, going deeper than a score — what the job really is, who the customer is, what the interview demands, and what trade is actually on the table. The output is a written decision that survives the session.
Advisory. Assess and draft in-message; do not apply, send, or create drafts unless asked.
1. Read, in this order — before writing anything
profile/rubric.md— the priority function. Authoritative over anything you remember, and the only copy that is current. Never judge a role without opening it.docs/knowledge-base/personal/roles/preferences.md— how the seeker wants to be advised: the screening-format calls, the settled questions, the recurring gaps. This is the anchor for this system.profile/bullet-bank.md— every fit claim traces here. Obey DO-NOT-CLAIM.docs/knowledge-base/personal/roles/— has this company been evaluated before? Filenames are<date>_<company>_<slug>.md.
Then check what the machine already knows, so you neither repeat it nor contradict it:
grep -ril "<company>" data/corpus/state-*.json | tail -2 # the triage record: score, red flags
ls data/research/<company-slug>.json # an existing company brief
grep -i "<company>" profile/skiplist.md # already applied or rejected?Read the triage record's analysis block, not a summary of it. A run that hit a provider error
writes a SKIP stub with fit_score: 0 and analysis_error in why — indistinguishable from a
judgment unless you look. If two runs disagree, the later one wins.
2. Get the posting itself
Read it to the end — comp, location, eligibility and the seniority bar are usually last.
If the source is an aggregator link, find the employer's own posting. Employers commonly open the same role at several grades and in several flavours (senior/non-senior, frontend/backend/full-stack), and the JDs are often identical apart from the years bar and the salary band. Always check the ladder — the lower rung may remove the only thing that would screen the seeker out.
curl -s "https://api.ashbyhq.com/posting-api/job-board/<slug>" # whole board + compensation
.venv/bin/python -m research --tool list_jobs "<company>" # or via the engine3. Research the company
Run /research-company (engine: python -m research "<company>"), then close its open questions with
your own web search and write them back with --answer.
Verify you have the right company. A board lookup resolves a name to an ATS slug and will happily land on a different company with the same name — check the employer's own careers URL, and treat review-site data as belonging to whichever company the site actually indexed.
4. The reply — one message, these sections
- What this is — company, channel, and whether it is the employer or an agency. Verified or inferred, with confidence.
- What the job actually is — the day-to-day, in plain terms. Correct the obvious misreading if there is one: which layer of the product, which team, what it is not.
- Who the customer is, when the role is customer-facing. It changes the job more than the stack does.
- Fit — each requirement, pass or miss, against the bank. Strong signals, honest gaps. Name what is underused as well as what is missing.
- The interview loop — find it; many employers publish one. Sizing it is part of the decision.
- Against the rubric — priority by priority, in the rubric's own order. Say which are for and which against. Held-back roles are shown with the reason, never hidden.
- The recommendation — answer the question that was asked. Surveying options without a pick is the failure mode.
- Open questions — what could not be verified, and what to ask on the first call.
Label every inference at the point of use. A guess stated as a fact is worse than no guess, and a hedge in paragraph one must not become a fact in paragraph five.
5. Record it
Write docs/knowledge-base/personal/roles/<date>_<company>_<role-slug>.md. Filenames match
applications/' <date>_<company>_<slug> so one grep finds a role in both places.
Carry the decision, the reasoning, the corrections, and the open questions — not the whole reply. The test of a good record: someone re-pitched this role in two months can read it and not redo the work.
Update it in place as a thread continues; do not open a second file for the same role.
When the decision is final, add the id to profile/skiplist.md with a one-line reason — that is the
machine index, and it is what stops the role resurfacing.
6. Propose preference updates — never write them
If the session produced a durable preference — a screening format that is a dealbreaker, a settled question, a gap that should always be handled the same way — propose it as one line plus where it came from, and let the seeker accept it.
Never edit preferences.md unprompted. Same precedence rule as the CV judge that grades but cannot
rewrite: a system that edits its own record of what someone wants is how a preference they never held
becomes permanent.
And never move a rubric value into it. preferences.md holds nothing profile/rubric.md can hold —
no rates, floors, tiers or scores. A real change to the priority function is a rubric edit.
Notes
- Not triage. This consumes triage's output; it does not re-run, re-score or edit it.
- Not a CV. When the decision is to apply, hand off to
/tailor-cv(one) or/tailor-cv-batch(several) and/cover-letter. - Everything under
docs/knowledge-base/personal/is pruned from the public snapshot, so company names, comp and personal circumstances belong there and nowhere one level up.
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.

