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.
npx agentmods add instructions/lycheeorg/lychee/agents-mdgit clone --depth 1 https://github.com/LycheeOrg/LycheeWhat 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 | $0.03757 | $0.03757 |
| Opus 5 | $0.01878 | $0.01878 |
| Sonnet 5 | $0.00751 | $0.00751 |
| Haiku 4.5 | $0.00376 | $0.00376 |
Grade C, and why
Lychee AGENTS.md scanned grade C with 1 finding 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- **No surprises.** Avoid destructive commands (e.g., `rm -rf`, `git reset --hard`) unless the user requests them. Stay within the repository sandbox. How it starts
The opening of the file, as written. The whole thing — 123 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Before You Code
- Clarify ambiguity first. Do not plan or implement until every requirement is understood. Ask the user, record unresolved items in docs/specs/4-architecture/open-questions.md, and wait for answers. Capture accepted answers by updating the relevant specification’s requirements/NFR/behaviour/telemetry sections so the spec remains the single source of truth for behaviour.
- No-direct-question rule: Never ask the user for clarification, approval, or a decision in chat until the matching open question is logged (table row + Question Details entry). Treat violations as blockers—stop work, add the missing entry, then resume the conversation by referencing that question ID.
- Whenever you present alternative approaches—whether for open questions or general solution proposals—first capture or update the entry in docs/specs/4-architecture/open-questions.md (summary row + Question Details section) so the ambiguity lives on disk, then present the stored text to the user inline in chat. Summarise the options directly in your reply (do not tell the user to open the file), call out the question ID (for example, Q013-01), and follow the numbered-heading/Options A, B, C format with pros/cons. Order the options by preference (Option A is always the recommended path, Option B the next-best, etc.) so the user sees our best advice first. Keep specifications/plans/tasks limited to that ID until the question is resolved.
- Work in small steps. During planning, break every change into logical, self-contained tasks that are expected to complete within ≤90 minutes. Execution can take longer if required; the goal is to plan manageable increments, and commit with a conventional message for each finished slice.
- Prime the knowledge map. Skim docs/specs/4-architecture/knowledge-map.md and the up-to-date module snapshot in docs/specs/architecture-graph.json before planning so new work reinforces the architectural relationships already captured there.
- Template usage. Author new specifications, feature plans, and task checklists using docs/specs/templates/feature-spec-template.md, docs/specs/templates/feature-plan-template.md, and docs/specs/templates/feature-tasks-template.md so structure, metadata, and verification notes stay uniform across features.
- ADR context. Before planning or implementation, skim ADRs under
docs/specs/5-decisionswhose related-features/specs entries reference the active feature ID so high-impact clarifications and architectural decisions are treated as required context alongside the roadmap, spec, plan, tasks, and knowledge map.
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.
- 2d ago First seen · 123 lines · 3,757 tokens per session scan C 12d64f7df591
Lychee AGENTS.md is an instructions file published in the GitHub repository LycheeOrg/Lychee (4,267 stars, last pushed today), licensed MIT. It adds 3,757 tokens to every session, about $0.0188 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
maui copilot-instructions.md
Guidance for GitHub Copilot when working on the .NET MAUI repository.
maui uitests.instructions.md
Instructions for dotnet/maui, covering ui testing guidelines for .net maui, ui test structure, two-project requirement, base class and infrastructure and naming conventions.
maui helix-device-tests.instructions.md
Guidelines for running and configuring .NET MAUI device tests on Helix infrastructure.
maui ci-copilot-pipeline-security.instructions.md
Security rules for the Copilot PR-review pipeline. Read before editing.
maui xaml-unittests.instructions.md
Guidelines for writing XAML unit tests in the Controls.Xaml.UnitTests project.
maui templates.instructions.md
Instructions for dotnet/maui, covering working with .net maui templates, 1. overview, 2. conditional compilation, build-time directives (platform-specific) and template-time directives (template parameters).