ctxrent
/ctxrent
Ocenia, ile kontekstu zużywają sesje agentów kodujących, i zamienia wnioski w reguły do CLAUDE.md lub AGENTS.md.
- Autor
- itsmylife44 ↗
- Repozytorium
- ctxrent ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 30.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 30.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Ocenia, ile kontekstu zużywają sesje agentów kodujących, i zamienia wnioski w reguły do CLAUDE.md lub AGENTS.md.
Oryginalny opis (EN)
Context rent of the user's coding-agent sessions, turned into approved rules for CLAUDE.md or AGENTS.md. Use when the user runs /ctxrent, asks where their tokens or context go, why sessions are slow or expensive, or wants hygiene rules from their own session history.
Jak działa
- 1. Get the command
- 2. Measure
- 3. Write the rules
- 4. Get approval
- Other requests
Podgląd SKILL.md
--- name: ctxrent description: Context rent of the user's coding-agent sessions, turned into approved rules for CLAUDE.md or AGENTS.md. Use when the user runs /ctxrent, asks where their tokens or context go, why sessions are slow or expensive, or wants hygiene rules from their own session history. license: MIT compatibility: Node 20 or newer. Reads Claude Code and Codex session records under the user's home directory. Runs the ctxrent CLI from PATH, or the latest release from npm through npx when it is not installed. metadata: version: "0.1.2" category: developer-tools allowed-tools: Bash(ctxrent *) Bash(npx -y ctxrent@latest *) --- # ctxrent Content kept in a context is re-sent on every request, so it costs its size times the turns it survives: its rent. `ctxrent` measures rent and returns findings. You turn the findings into rules; the user approves them on a page. Run the whole flow yourself, start to finish; the only thing the user does is tick and press Apply. If the user asked for something narrower (a gate, a shareable report, a comparison), skip to [Other requests](#other-requests). ## 1. Get the command If `ctxrent --help` succeeds, the command is `ctxrent`. Otherwise it is: ```bash npx -y ctxrent@latest ``` That runs the latest release from npm without installing anything; npx checks for a newer one on every run. Below, `ctxrent` means whichever one works. Done when `ctxrent --help` prints the usage. ## 2. Measure ```bash ctxrent findings --json ``` Narrow it if the user asked: `--since <date>`, a session path as the first argument, `--min-sessions <n>` (default 3). Each finding has `id`, `label` (what was observed, as data), `lever`, `saving` (token-turns), `claims`, `sessions` and `example` (the real command that caused it). If `findings` is empty, run `ctxrent` alone, show the user the last session's rent, score and top carriers, and stop: there is nothing recurring to write a rule for. If the target (see step 4) has a managed block that says `applied <date>`, judge the rules already there: run `ctxrent check --all --since <date> --max-waste 100 --json` and the same without `--since`, and tell the user the waste share (`actual`) since the rules against all sessions. Exit 2 with `--since` means no session since then: say it is too early to tell. Done when you hold the JSON, and the comparison when the block has a date. ## 3. Write the rules One rule per finding whose `lever` is `instruction`. A `configuration` finding gets no rule: it is caused by a hook or a setting, and the page shows it to the user as such. Write them to `ctxrent-rules.md` in the system temp directory, outside the user's project, one per line, exactly: ``` - <the rule> <!-- id: <finding id> --> ``` Each rule is imperative, addressed to the agent, at most 20 words, in the language the user writes in. Decide what it says from `label`, `example` and the numbers. Leave a finding out when you are not confident about its rule: a missing rule beats a wrong one. Copy ids from the JSON only; `ctxrent` drops any line whose id it does not know. When the target already has a rule for a finding, write a new one only if it should change: the page offers it as a rewrite, unticked, and the current wording stays unless the user ticks it. Done when every `instruction` finding has one line or was left out on purpose, and the file holds nothing else. ## 4. Get approval The target is the global instruction file of the harness you run in: `~/.claude/CLAUDE.md` for Claude Code, `~/.codex/AGENTS.md` for Codex, or the file the user named. ```bash ctxrent review --rules <rules file> --write <target> ``` It prints a loopback URL, opens the browser, and waits up to ten minutes for the user to press Apply. Run it in the background or with a timeout of at least ten minutes, because a shorter timeout kills the page. As soon as the URL is printed, give it to the user with one line per finding: the amount, the example, and for `configuration` findings the hook or setting to turn off. The page is the only writer of the instruction file. Your edits stop at the rules file. Done when the command exits. Report its last line: `applied N rule(s) to <file>`, or `closed without changes`. ## Other requests - A gate: `ctxrent check --max-waste <n>`, or `--all --since <date>` for the archive. Exit 0 under the threshold, 1 over, 2 when the record cannot be measured. - A shareable report: `ctxrent audit <path> --html <file>`, one self-contained file. - A comparison of two variants: `ctxrent init`, then `ctxrent diff ctxrent.json --a <dir> --b <dir>`.
Wersje
W bazie nie ma innych skilli o nazwie ctxrent.
Migawka bazy z 13.11.2026 · wersja paczki 9d73d7ba0de8 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100context-rent
Wskazuje, co zjada okno kontekstu i budżet tokenów w sesjach Claude Code i Codex, i pomaga to ograniczyć. - 100context-cost
Mierzy, ile tokenów agent czyta przed pierwszym pytaniem: łańcuchy CLAUDE.md i AGENTS.md, pliki pamięci, reguły, umiejętności i serwery MCP. - 100lean-context
Utrzymuje tanie sesje agentów kodujących w dużych projektach z wieloma repozytoriami: CLAUDE.md, mapa projektu, ustawienia i narzędzia. - 86setup
Instaluje i podłącza memry, trwałą pamięć dla agentów kodujących, gdy użytkownik chce, by Claude Code pamiętał kontekst. - 100harness-creator
Buduje, audytuje i poprawia harnessy dla agentów kodujących: pliki AGENTS.md i CLAUDE.md, śledzenie stanu, bramki weryfikacji, granice zakresu, przekazania sesji, pamięć, budżety kontekstu i koordynację wielu agentów. - 100coding-agent-sessions
Znajduje, czyta i odtwarza sesje agentów kodujących z logów Codex, Claude, OpenCode, OMO/Senpi i innych, także transkrypty i uruchomienia subagentów.