agentleFS
AcKeskinAcKeskin/contexture/skills/deliver/SKILL.md

deliver

Skill2 starsChanged 7 days agoMITView on GitHub

Render a selection of retrieved fragments into Claude's working context per the deliver format contract. Library-only — callable by other skills (discover, prep, review, retrospect-core) that need to present bodies, not a user-invoked skill. Stateless.

25 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use prune, create-issue and 8 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • matthewpSkill

    prune

    Find and remove unused CSS in a Tideway project — classes defined in stylesheets but never referenced in markup. Use whenever the user wants to prune dead CSS, find unused styles or classes, clean up or slim a stylesheet, find orphaned BEM classes, or "tree-shake" hand-written CSS.

    Also in 3 projects that use this

  • sahinSkill

    create-issue

    Also in 3 projects that use this

  • BlockSkill

    plan

    Draft a TDD, PRD, or lightweight plan in the conversation without committing it.

    Also in 5 projects that use this

  • AWSSkill

    ship

    Finalize dev work on an issue branch — update docs, push, and open a PR.

    Also in 5 projects that use this

  • AWSSkill

    wait-what

    Stop. That last message did not land: re-pitch it.

    Also in 3 projects that use this

  • Hugging FaceSkill

    implement

    Make tests pass. Invoke after /write-tests produces failing tests.

    Also in 5 projects that use this

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 25 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Default priority (highest tier first):

    line 54
  2. Very specific:

    If order_override == "caller", skip tiers and respect the caller's exact sequence.

    line 65
  3. Distinctive:

    If a single tier's members must be split, keep the head of that tier in caller-given order.

    line 72
  4. Distinctive:

    If fragments were dropped in step 3, append exactly one trailing line after the last fragment:

    line 100
  5. Distinctive:

    Does not maintain state: No cache, no history, no memoization.

    line 122

“Common” is measured: kept by at least 6 in 10 of the 25 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

deliver

The delivery organ. Takes an already-selected set of fragments and produces the rendered block Claude reads.

When to run

  • Another skill invokes this one programmatically, passing a fragment selection it has already filtered and scored (discovery for bodies, prep at task start, review during diff inspection, plan at milestone boundaries).
  • Do not respond to user prose triggers. There is no /deliver command. There is no "when user says X" branch. Delivery is purely a format-and-assembly helper.
  • Stateless per call. No memoization, no cache, no hidden state.

Inputs

Structured, from the caller:

{
  fragments: Fragment[],        // already-selected; delivery does not re-filter
  order_override?: Order,       // optional; default = the §2 tier priority
  cap?: number,                 // optional; default 20; hard ceiling regardless of caller
}

Fragment = {
  path: string,                 // repo-relative or ~/-relative; source-of-truth pointer
  frontmatter: {                // parsed YAML
    name: string,
    description?: string,
    type?: string,
    kind?: string,              // "architectural-rule" | "decision" | "lesson" | "preference" | ...
    scope?: string[],
    relevance?: string,
  },
  body: string,                 // verbatim content — delivery does not paraphrase
}

Order = "default" | "caller" | explicit-tier-list

Procedure

1. Validate inputs

  • fragments present and non-empty. Empty → return empty string and a single-line note ([delivery: empty selection]). Never synthesise content.
  • Each fragment has path, frontmatter.name, body. Missing path or name → render with the available fields, emit a [delivery: fragment missing <field>] marker inline so the caller notices.
  • cap bounded to [1, 20]. Caller-specified values outside this range are clamped; flag the clamp in the report line.

2. Sort

Default priority (highest tier first):

  1. Warnings — frontmatter.kind == "warning". Landmines that earned their place by citing a real prior incident; surface ahead of everything because the cost of forgetting one is concrete.
  2. Architectural rules — frontmatter.kind == "architectural-rule".
  3. Project-specific facts — frontmatter.type == "project" OR frontmatter.scope contains any project-* tag.
  4. Session recaps — frontmatter.type == "session-recap". Episodic recall sits below project facts but above codemap, because "what I was investigating" gives more signal than raw file summaries when resuming work.
  5. Codemap entries — fragments where frontmatter.kind == "codemap" or frontmatter.type == "codemap". (Discovery surfaces these when .claude/codemap.md contributes.)
  6. Everything else — lessons, decisions, preferences, references, general feedback.

Within each tier, preserve caller's ordering (stable sort — typically discovery's score descending).

If order_override == "caller", skip tiers and respect the caller's exact sequence. If order_override is an explicit tier list, use that in place of the default.

3. Cap

Hard cap is the resolved cap value (default 20). When fragments.length > cap:

  • Drop from the lowest-priority tier first.
  • If a single tier's members must be split, keep the head of that tier in caller-given order; drop the tail.
  • Record the count of dropped fragments.

4. Render

Per-fragment shape, exact:

[scope: <comma-joined scope tags> | kind: <kind or "lesson"> | path: <path>] <name>
<body verbatim>

For kind: warning, prepend a callout line before the header so the warning is visually distinct from neighbors:

📛 WARNING — do not repeat
[scope: <...> | kind: warning | path: <...>] <name>
<body verbatim>

Notes:

  • scope joins frontmatter.scope with , . If empty → [scope: -].
  • kind defaults to lesson when absent (the memory-format default).
  • path verbatim from input — no normalisation, no resolution.
  • body verbatim — no trimming beyond preserving the original content, except strip architectural-rule bullet-id anchors (<!-- id: ... --> prefixes): they are patch-resolution targets, never context. This is the one sanctioned body edit — it removes machinery, not content. Discover's resolver strips them too; deliver does it as defense-in-depth so a caller that bypasses the resolver still never leaks anchors. No other reformatting.
  • One blank line between fragments.
  • Warning callout is the only prepend — other kinds render plain. Keep the callout to one line; the body itself carries the actual content.

If fragments were dropped in step 3, append exactly one trailing line after the last fragment:

[N fragments dropped — narrow scope for more detail]

5. Return

Return the rendered string to the caller. Caller owns display decisions (show to user, feed to another skill, etc.). Delivery does not log, does not echo, does not persist.

Failure modes

  • Empty selection. Return empty string + single-line [delivery: empty selection] marker. Caller decides whether to suppress.
  • Malformed fragment (missing path or name). Render what's available; inline a [delivery: fragment missing <field>] marker above the fragment. Never skip silently.
  • Body longer than reasonable (> ~5KB or so). Render verbatim anyway — delivery is not a censor, and compression is storage-time, not delivery-time. If the caller passes a giant fragment, that's a caller problem.
  • order_override unrecognised value. Fall back to default priority and emit [delivery: unknown order override, fell back to default].

What deliver does NOT do

  • Does not select fragments. Selection / filtering / scoring belongs to discover or the calling skill.
  • Does not compress, paraphrase, or summarise. Compression is storage-time. Delivery is faithful to source.
  • Does not enforce relevance. Bodies passed in are assumed relevant by the caller.
  • Does not maintain state. No cache, no history, no memoization.
  • Does not display to the user. Caller decides whether the rendered block is shown, stored, or consumed internally.
  • Does not modify or re-read source files. Operates only on the passed-in fragment structs.

Source-of-truth rule

Every rendered fragment carries its path in the header. Callers and downstream readers (Claude re-reading in the same session, or the user inspecting the block) can always trace a fragment back to its authoritative file. Delivery never rewrites a fragment — the rendered form is a presentation of the original, not a replacement.

See docs/delivery-organ.md for the scope map and the rationale for each rule.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.