Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/Xclaw-bot/benchmark-task-authoringnpx agentmods add rules/xclaw-bot/benchmark-task-authoring/05-retrievalWrote 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/rules/xclaw-bot/benchmark-task-authoring/05-retrieval)<a href="https://agentmods.dev/rules/xclaw-bot/benchmark-task-authoring/05-retrieval"><img src="https://agentmods.dev/badge/rules/xclaw-bot/benchmark-task-authoring/05-retrieval/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/rules/xclaw-bot/benchmark-task-authoring/05-retrieval"><img src="https://agentmods.dev/badge/rules/xclaw-bot/benchmark-task-authoring/05-retrieval.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 | $0.00546 | $0.00546 |
| Opus 5.5 | $0.00218 | $0.00218 |
| Sonnet 5.5 | $0.00109 | $0.00109 |
| Haiku 4.5 | $0.00055 | $0.00055 |
Grade A, and why
05-retrieval 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 2mo 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 — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
description: Retrieve from the local index instead of re-reading long documents. alwaysApply: true
Retrieve, do not re-read
The the corpus is ~193k tokens across 49 files. Re-reading a war-chest to re-extract a law is the single largest avoidable token cost in this workspace, and it scales with every slot shipped.
Before opening any of ../hardness-laws.md, a <task-repo>.md memory
file, ../hardness-laws.md, or a past proposal-*.md — query the index.
python scripts/dr.py boot # session start (~3k tok, replaces ~18k)
python scripts/dr.py ask "<question>" # ~1-2k tok, <1s, cited
python scripts/dr.py ask "<q>" --task <hash> --klass law --budget 1200
python scripts/dr.py check <proposal.md> # laws a draft is likely breaking
python scripts/dr.py laws --write # regenerate the distilled deck
Every returned card cites a real file:line (verified 40/40). Opening that file
to read around a card you were given is a targeted read and is fine. Opening a
file to search it is the thing to stop doing.
Run dr.py index after any memory write — incremental, ~1s.
Cache measurements, never re-derive or re-run them
Standing rule: never fabricate a measurement. The cache is how you keep that
rule cheaply — a stored result is fingerprinted against the tree that produced
it, and get fails with exit 4 if that tree has changed since.
# store: pipe the command's output straight in
python preflight.py <task-dir> 2>&1 | \
python scripts/dr.py cache put preflight-<hash> --stdin --watch <task-dir>
# reuse, or re-run if missing (3) or stale (4)
python scripts/dr.py cache get preflight-<hash> --watch <task-dir> \
|| python preflight.py <task-dir>
python scripts/dr.py cache list
Cache preflight output, probe solve rates, timeout-probe wall clocks, oracle/nop
rewards. A STALE result is a signal to re-run, never to quote the old number.
Always pass --watch: without it nothing is verified, and the tool will say so.
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.
- 2mo ago First seen · 51 lines · 546 tokens per session scan A 8a9bac59b1eb
05-retrieval is a cursor rule published in the GitHub repository Xclaw-bot/benchmark-task-authoring (2 stars, last pushed 2mo ago), licensed MIT. It adds 546 tokens to every session, about $0.0022 per session on Opus 5.5. 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.
Other cursor rules, from other repositories
mempalace-recall-always
Always-on MemPalace recall — search the palace before answering about past work, people, projects, or prior decisions.
dreamd-recall
Search dreamd at the start of a task and quote a hit. Append when the user corrects you or states a decision.
session-memory
Use at conversation wrap-up or when the user explicitly indicates end-of-session — capture residual lessons not captured in-flight.
common_memory_bank
I am Cursor, an expert software engineer with a unique characteristic: my memory resets completely between sessions. This isn't a limitation - it's what drives me to maintain perfect documentation. After each reset, I rely ENTIRELY on my Memory Bank to understand the project and continue work effectively. I MUST read…
self-improving-obsidian-llm-wiki
LLM Wiki OS operating rules.
memsprout
Cursor rule "memsprout" from memsprout/agents, covering using memsprout, search first, capturing, spaces and topics and files.