Context Compactor

Condense a long agent session transcript into a compact handoff memo: decisions, tasks, risks, facts & links. Credential values are auto-redacted before output (best-effort - review memos before sharing; --strict drops whole suspect lines). EN/RU heuristics, zero dependencies. The CLI prints what it reads and where it writes. Use ONLY with the user's explicit consent: tell the user which transcript file will be read.

Local-first. No telemetry, no cloud sync — the transcript you feed it never leaves your machine.

When to use

• Your session is about to be compacted and you want a handoff memo to paste into the next one.
• You hand work to another agent (or a human) without dumping 500 KB of transcript.
• You want a daily digest of what was actually decided and what is still open. Don't use for: summarizing code repositories, translating documents, chat-style Q&A.

Quick start

bash

# from this package (or the git repo: github.com/vnbochkarev-netizen/context-compactor)
python3 compactor.py --input session.md
python3 compactor.py --input session.md --output handoff.md
python3 compactor.py --self-test

Safety first: credential values are redacted (best-effort)

Handoff memos travel to other agents and files. Context Compactor redacts credential values before scoring: password, token, api key, secret, sk-…/ghp_… tokens, JWTs, URL-embedded credentials and PEM private-key blocks become <REDACTED: …>. The surrounding decision/context is kept — the secret value is not.

⚠️ Redaction is best-effort pattern matching — review memos before sharing them. --strict drops whole lines that still carry a credential marker. --self-test asserts the covered cases (incl. URL creds, JWT, Cyrillic, PEM > 600 chars).

How it works

• Bucketing: strong decision verbs (EN and RU, e.g. "decided") → Decisions; weak markers with a link ("agreed, link: ") → Facts; tasks, risks and links get their own section.
• Long transcripts: the head AND the tail are analyzed (final decisions usually sit at the end); the skipped middle is counted and reported in the memo header together with the tool version, date and source file.
• Code dumps and tool noise are filtered out; duplicates are removed.

Privacy, consent, retention & deletion

What	Where	How to delete
Transcript file you point at	in memory only	nothing is written
Output memo	stdout or --output <file> you choose	delete the file

Get explicit consent before processing a transcript: tell the user which file will be read. The tool writes nothing by itself, phones nothing and keeps no logs.

Permissions

• Files: read-only access to the transcript file the user explicitly points at; writes only to an --output path the user provides.
• Process: none — no subprocesses, no installs.
• Network: NONE.
• Secrets: values are redacted before any output; the tool never stores or transmits them.

License

MIT © 2026 Viacheslav Bochkarev. Free to use, modify and redistribute.