/**
 * Centralized skill system limits.
 *
 * Before this file existed, the magic numbers below were scattered across five
 * modules (`skill-installer`, `github-fetcher`, `skill` tool, system-prompt
 * builder, skill-loader), sometimes duplicated (e.g. per-skill body cap was
 * defined independently in the builder and the tool). Centralizing them here
 * keeps the budget model coherent — change one, change everywhere — and makes
 * the trade-offs visible in one place.
 *
 * Values are frozen at their pre-centralization defaults so this is a pure
 * refactor (no behavior change).
 */
export declare const SKILL_LIMITS: {
    /**
     * Per-skill body cap when injecting a full skill body into the system prompt
     * (eager mode), and when returning a skill body from the `skill` tool.
     * ~4k tokens. Oversized bodies are truncated at a paragraph boundary.
     *
     * Consumers: `system-prompt-builder.capSkillBody`, `skill` tool body return.
     */
    readonly MAX_SKILL_BODY_CHARS: 16000;
    /**
     * Per-resource cap when the `skill` tool loads a bundled file
     * (`scripts/`, `references/`, `assets/`). ~8k tokens.
     *
     * Consumer: `skill` tool `loadResource`.
     */
    readonly MAX_RESOURCE_CHARS: 32000;
    /**
     * Maximum number of bundled resource paths the `skill` tool lists in one
     * response. Caps a pathological skill (huge `assets/` tree) from blowing up
     * the tool result.
     *
     * Consumer: `skill` tool resource listing.
     */
    readonly MAX_LISTED_RESOURCES: 100;
    /**
     * Max size of a single installed skill file (SKILL.md or a bundled resource).
     * Guards against a malicious registry skill shipping a multi-MB blob that
     * then flows into the prompt. 100KB.
     *
     * Consumer: `SkillInstaller.install` / `importFromDir`.
     */
    readonly MAX_SKILL_FILE_SIZE: number;
    /**
     * Max size of a GitHub tarball downloaded by the skill installer. Guards
     * against a repo accidentally (or maliciously) shipping a giant tarball that
     * would be extracted into a temp dir. 50MB.
     *
     * Consumer: `downloadGitHubTarball`.
     */
    readonly MAX_TARBALL_SIZE: number;
    /**
     * Default total char budget for skill bodies injected in eager mode when
     * `config.skills.eagerMaxChars` is unset. ~6k tokens. Skills are injected
     * highest-priority first; once the budget is exhausted the remaining skills
     * are listed as a manifest the agent loads via the `skill` tool. Set very
     * high to effectively disable budgeting.
     *
     * Consumer: `DefaultSystemPromptBuilder.buildMemoryAndSkills` (default for
     * `skillEagerMaxChars`).
     */
    readonly EAGER_DEFAULT_MAX_CHARS: 24000;
    /**
     * Auto-compact body limits (the token-saving fallback used when a skill has
     * no hand-crafted `SKILL.save.md`). The Overview and Rules sections are
     * extracted and trimmed to these char budgets, then the total is capped.
     *
     * Consumer: `DefaultSkillLoader.compactSkillBody`.
     */
    readonly COMPACT_OVERVIEW_MAX: 200;
    readonly COMPACT_RULES_MAX: 350;
    readonly COMPACT_TOTAL_MAX: 450;
    /**
     * Max length of a skill name (agentskills.io spec). Enforced by the
     * frontmatter validator's format regex plus this length bound.
     *
     * Consumer: `isValidSkillNameFormat`.
     */
    readonly SKILL_NAME_MAX_LEN: 64;
    /**
     * Soft line limit for a SKILL.md body, per the bundled `skill-creator` rule.
     * Not enforced in code (a skill can exceed it) — surfaced by `/skill-gen
     * validate` and the `skill-creator` skill as guidance to move deep material
     * into `references/`.
     *
     * Consumer: `skill-creator` skill, `/skill-gen validate` advisory.
     */
    readonly SKILL_BODY_LINE_LIMIT: 500;
};
//# sourceMappingURL=limits.d.ts.map