distill

A command that extracts reusable knowledge from a work session and stores it in the project profile.

In plain words
What is it for?
Use it to record durable project knowledge, including evidence and user-provided facts, in the project’s profile files.
Why use it?
It preserves decisions and facts that can help with later work across different features.

Command for Claude Code

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 commands/zts0hg/codexspec/distill
Clone the repo
git clone --depth 1 https://github.com/Zts0hg/codexspec

Made for: Claude Code.

Per session 14 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 4,966 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 $0.00014 $0.04966
Opus 5 $0.00007 $0.02483
Sonnet 5 $0.00003 $0.00993
Haiku 4.5 $0.00001 $0.00497

Measured 3d ago against content hash 1b4ad88e050d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

distill 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 3d 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.

.claude/commands/codexspec/distill.md · 217 lines

How it starts

The opening of the file, as written. The whole thing — 217 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Distill

Language Preference

Read .codexspec/config.yml. Two independent language controls apply (each falls back to language.output, then English):

  • Interaction language (language.interaction): language for all conversation with the user — questions, explanations, status messages, and codexspec CLI terminal output.
  • Document language (language.document): language for generated artifact files (the profile records).

Converse in the interaction language and author artifacts in the document language. Apply the project's translation standard to both: translate by meaning (not word-for-word), keep English for terms with no good native equivalent, and write as if originally in that language. Exception: evidence.facts quotes the user's original words verbatim and MUST NOT be translated.

User Input

$ARGUMENTS

Operating Model

distill extracts the reusable, cross-feature knowledge produced during work and persists it to the project-level store .codexspec/profile/. It runs two ways:

  • Auto (primary): embedded in wrap-up commands (implement-tasks on completion, commit-staged, pr), gated by workflow.auto_distill in .codexspec/config.yml (default enabled; disabled only when explicitly set to the literal false).
  • Near-moment (ambient): driven by the profile block's "capture knowledge as you go" rule, distill may be invoked near the moment reusable cross-feature knowledge is produced — in any session, including plain chat or a non-SDD fix that never reaches a wrap-up command. This is the primary way knowledge from ad-hoc work is captured at all.
  • Long-run: in a long-running implement-tasks, distill along the way near each knowledge-producing event rather than only at the very end, so mid-task evidence is not lost to context compaction; the end-of-task auto_distill still runs as a backstop.
  • Manual (fallback): invoked directly on the supplied or most-recent interaction segment.

Read the full file on GitHub · 217 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. 3d ago First seen · 217 lines · 14 tokens per session scan A 1b4ad88e050d

Subscribe to this mod's changes

distill is a command published in the GitHub repository Zts0hg/codexspec (5 stars, last pushed 4d ago), licensed MIT. It adds 14 tokens to every session and 4,966 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 2026-08-31.