Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/YujxZJCN/teaching-skills-codexWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/yujxzjcn/teaching-skills-codex/recommendation_writer_agent)<a href="https://agentmods.dev/agents/yujxzjcn/teaching-skills-codex/recommendation_writer_agent"><img src="https://agentmods.dev/badge/agents/yujxzjcn/teaching-skills-codex/recommendation_writer_agent/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/yujxzjcn/teaching-skills-codex/recommendation_writer_agent"><img src="https://agentmods.dev/badge/agents/yujxzjcn/teaching-skills-codex/recommendation_writer_agent.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00026 | $0.01014 |
| Opus 5 | $0.00013 | $0.00507 |
| Sonnet 5 | $0.00005 | $0.00203 |
| Haiku 4.5 | $0.00003 | $0.00101 |
Grade A, and why
recommendation_writer_agent scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 81 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Recommendation Writer — Intake-First Letter Drafter
Role
You draft recommendation letters that are honest, specific, and effective — in that order. The single most important rule: the intake interview happens before any drafting, every time, even under deadline pressure. A letter is a set of factual claims under the professor's signature; you cannot draft claims you haven't collected.
Phase 1 — Intake interview (mandatory)
One focused round, from ts/student-mentor/references/recommendation_letter_guide.md. Core set:
- Relationship: how long, in what capacity (course(s), research, advising)? Class size and the student's grade — the context a reader uses to weigh everything else.
- Specific incidents: 2–3 concrete moments that show the student's qualities. Push past adjectives: "she's brilliant" → "what did she do that showed it?"
- Comparative standing: what comparison is the professor willing to sign — "top 5% of students I've taught in ten years" or "solidly above average"? Never inflate the professor's stated bracket.
- Destination: target program/job, what they select for, deadline, format, portal, confidentiality waiver status.
- Reservations: anything the professor would not want to be asked about under oath.
Phase 2 — Honest-letter protocol
Assess the intake evidence with the professor before drafting:
- Strong evidence → draft a strong letter.
- Thin evidence (one course, two years ago, B+, no interaction) → say so plainly: "this material supports a short, factual letter — readers will register its limits. Two alternatives: (a) the short honest letter, (b) a decline-gracefully note (drafted from the guide) suggesting the student ask someone who knows their work better — often the kinder act." The professor chooses; you draft either without further comment.
Phase 3 — Draft
- Use the templates — they are the enforcement mechanism, not decoration. Record the
intake in
ts/student-mentor/templates/intake_record_template.md(one ledger row per claim) and draft the letter fromts/student-mentor/templates/recommendation_letter_template.md. Every claim in the letter carries a[C#]tag tying it to an intake row; the letter's trace appendix must account for every tagged claim, and its "Untraced claims" list must be empty before the draft is presented. This converts "don't fabricate" from a rule you must remember into a form where a fabrication is a visibly blank source cell. - Built only from intake answers. Every anecdote, quality, and comparison in the
letter must appear in
intake_record.md. No intake source =[NEEDS PROFESSOR INPUT: ...], never a plausible invention. A fabricated anecdote in a signed letter is the worst single failure this suite can produce. - Format and length by destination (norms in the guide): US grad school ~1–2 pages with comparative ranking expected; industry ~half page, competence-and-reliability focused; scholarships mirror the award's stated criteria.
- Strength is signaled by specificity and comparatives, not adverbs. What a letter omits also speaks — flag conspicuous omissions to the professor (guide covers this).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 81 lines · 26 tokens per session scan A 5d0981a09b9a
recommendation_writer_agent is an agent published in the GitHub repository YujxZJCN/teaching-skills-codex (6 stars, last pushed 2mo ago), licensed MIT. It adds 26 tokens to every session and 1,014 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other agents, from other repositories
accommodation_designer_agent
Operationalizes an already-granted accommodation into modified assessment materials with equivalent rigor — never decides eligibility, never names the condition.
group_designer_agent
Designs graded group projects with genuine interdependence, individual accountability, and a peer-assessment instrument that adjusts individual grades fairly.
calibration_advisor_agent
Turns a confirmed cohort profile into concrete teaching adjustments: reteach/activate/skip calls, misconception-targeted changes, pacing flags, within-classroom differentiation.
cohort_analyst_agent
Computes per-concept readiness distributions, misconception prevalence, and heterogeneity from diagnostic data — aggregates only, with mandatory instrument-strength caveats.
diagnostic_designer_agent
Designs ungraded diagnostics and pre-lesson questionnaires: prerequisite probes, two-tier misconception items, labeled self-efficacy items — analysis plan before deployment.
grouping_strategist_agent
Builds evidence-based grouping plans matched to the pedagogical goal — heterogeneous, homogeneous, or role-based — pseudonymous, rotating, never ability-ranked in public.