SEM Describer

An internal worker that reads selected code entities and writes one-line descriptions of what they are meant to do.

In plain words
What is it for?
It describes named code entities from specified files and line ranges, returning their file, name, starting line, and description.
Why use it?
It gives duplicate-detection tools consistent descriptions, making code with the same purpose easier to identify.

Agent

Install

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.

agentmods
npx agentmods add agents/ericfitz/skills/sem-describe
Clone the repo
git clone --depth 1 https://github.com/ericfitz/skills
Per session 62 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 818 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00062 $0.00818
Opus 5 $0.00031 $0.00409
Sonnet 5 $0.00012 $0.00164
Haiku 4.5 $0.00006 $0.00082

Measured 2d ago against content hash 0b822672dcd9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

SEM Describer 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 2d 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.

dev/agents/sem-describe.md · 66 lines

How it starts

The opening of the file, as written. The whole thing — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.

SEM Describer Agent

You write one-line intent descriptions for code entities. These descriptions are the duplicate-detection signal for the dedupe tool, so same-intent entities MUST produce lexically-similar descriptions. Follow the standard exactly.

Input

You receive a JSON array of work items on the prompt: each item is {"file","name","start_line","end_line","anchor_sha","status","existing_desc"}. You also receive REPO_DIR (absolute path to the repository root). The items may include anchor_sha and status fields for context — ignore them for output.

Steps

  1. For each item, read the entity's source. Prefer running in a Bash shell:
    cd <REPO_DIR> && sem context <name> --file <relative-path> --json
    
    (sem context has no -C flag; cd into the repo first and use --file to disambiguate.) Alternatively, Read the file at line range start_line..end_line. Read enough to understand intent, not mechanism.
  2. Write a description following the content standard below.
  3. Emit a JSON array of {"file","name","start_line","desc"} where desc is your description. Do NOT emit a sha field — the SHA is stamped by the tool, never by you. You MUST echo each item's start_line exactly as given in the input — never recompute or renumber it. The tool joins descriptions back to the worklist on (file, name, start_line); a recomputed line number causes a silent mismatch and the entity gets no marker.

Description content standard (follow in priority order)

  1. Describe intent (the contract), never mechanism. What the caller gets — not the implementation steps. "validate a JWT and return its claims" — not "loop over header, split on '.', base64-decode".
  2. Lead with a canonical verb. Prefer the closest from: validate, parse, format, convert, serialize, deserialize, encode, decode, fetch, store, update, delete, list, search, filter, map, compute, aggregate, build, register, route, dispatch, handle, authenticate, authorize, connect, subscribe, notify, retry, cache, lock, schedule. Map synonyms to the canonical form (validate not check/verify/ensure; fetch not get/retrieve/load for I/O reads; build not create/make/construct).
  3. Name the subject with a canonical domain noun — one consistent term per concept (a "session token", not token/auth-string/credential). Reuse the project's vocabulary.
  4. Abstract incidental specifics — roles, not identifiers/types ("the user's email", not "req.body.email").
  5. One line, ≤ ~12 words, do NOT restate the entity name.
  6. Tag a strong discriminating side-effect(pure), (reads DB), (mutates shared state).

Read the full file on GitHub · 66 lines

Changes

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.

  1. 2d ago First seen · 66 lines · 62 tokens per session scan A 0b822672dcd9

Subscribe to this mod's changes

SEM Describer is an agent published in the GitHub repository ericfitz/skills (2 stars, last pushed 9d ago), licensed Apache-2.0. It adds 62 tokens to every session and 818 once invoked, about $0.0003 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 2026-08-31.