@firefly0621/dsh-skill-always-apply

English | 中文

Opt-in Cordis consumer that injects skill bodies marked alwaysApply: true (Cursor-aligned frontmatter) into a session before the first model request, without requiring a skill tool load.

This package does not register a skill provider. Mount it beside dsh-skill / dsh-skill-filesystem (and usually dsh-tool-skill). It is not part of the default dsh-base composition; install with:

dsh plugin --profile web add @firefly0621/dsh-skill-always-apply

Or insert the shipped cordis.patch.yml into a custom base.

The package also runs against upstream @deepseek-ai/dsh-skill releases that predate the typed alwaysApply field. When a catalog summary does not carry the field, this consumer loads the candidate and reads alwaysApply from the definition or the skill file's frontmatter, so no host-code change is required.

Skill frontmatter

In a local SKILL.md (or flat .md skill) parsed by @deepseek-ai/dsh-skill-filesystem:

---
name: my-standing-rules
description: Standing session rules
alwaysApply: true
---

Follow these rules for the whole session.

alwaysApply uses the same boolean spellings as disable-model-invocation / user-invocable. An invalid value is treated as not opted in, so this consumer skips injection while the skill stays in the discovery catalog.

Routing copy (description / whenToUse) stays on the skill provider that registers the skill; this consumer only selects and injects bodies. Mount skill providers that publish clear routing text so discovery and the model catalog stay useful.

The shipped @firefly0621/dsh-skill-karpathy-guidelines provider marks karpathy-guidelines with alwaysApply: true. Mount this consumer next to that provider to inject the Karpathy body without a skill tool load.

Behavior

On the first applicable agent/pre-step (after nested listeners such as the skill catalog):

1. Skip when session.header.origin === 'subagent' (default; skipSubagent: false overrides).
2. Skip when the current step already carries a skill-always-apply message, or when one remains on the model-visible surface.
3. snapshot() the viewing agent's skills; skip incomplete observations.
4. Select Config.names plus skills whose summary, loaded definition, or skill-file frontmatter carries alwaysApply: true, minus disabledNames.
5. Load each body via ctx.skills.get(), honor complete-message maxTotalBytes, and prepend one durable user-role instructions message that lists names and embeds each renderSkillContent block.

Resume sessions still showing the message on the surface are not re-injected; surface-shadowed messages are.

Config

Always-apply injection is a host standing-instructions path: frontmatter alwaysApply: true and Config.names do not require modelInvocable. A skill with disable-model-invocation: true can still be injected here while staying out of the model-facing skill catalog.

Model Experience

Always-apply standing instructions

What the model sees

One durable user-role instructions message before the first model request when at least one selected skill fits the complete-message budget: a short <system-reminder> naming the always-apply set, then each skill's canonical <skill_content> block. Later steps reuse the same session history while the message remains on the model-visible surface. If compaction (or another surface replace) shadows the message, the next pre-step re-injects.

Token effect

One retained instructions message whose size is the reminder envelope plus every rendered skill body that fit under maxTotalBytes.

KV Cache effect

While the injection stays visible, it is append-once. Surface-shadowed re-injection appends a new durable copy.

Known Limitations and Deferred Work

• Opt-in only — product defaults do not mount this plugin; operators add it explicitly.
• No membership refresh while visible — after a visible injection, newly appearing always-apply skills are not appended until a new session or a surface-shadowed re-inject rebuilds the set from the current catalog.
• Catalog still lists model-invocable always-apply skills — those entries remain in the skill catalog; the reminder tells the model not to re-load them when the body is already present.
• Bypasses model invocation policy — always-apply and Config.names inject regardless of modelInvocable; use disabledNames to exclude.