agentleFS
mhxiemhxie/Atelier/skills/hi/SKILL.md

hi

Skill5 starsChanged 35 days agoMITView on GitHub

Universal entry point — intent router for reflection, planning, action, reading, learning, capture, and more.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    When /hi or $hi has no following text, do not load the catalog.

    line 16
  2. Very specific:

    Run uv run scripts/intent_coverage.py catalog --examples and read it: one line per harness/intents.toml row (plus any private rows from the …

    line 22
  3. Distinctive:

    Do not clarify a request whose reading is obvious in context.

    line 36
  4. Very specific:

    When the row declares profile_reads, run:

    line 61
  5. Very specific:

    Read only the row's procedure file and execute it as the selected workflow.

    line 73
  6. Distinctive:

    When the row is parallel, dispatch the declared initial agents in one native batch.

    line 74

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

Outcome: select one intent, load only its bounded context and procedure, and execute it through the active runtime. Done when: the route is visible, real ambiguity is resolved, and the selected procedure's completion criteria are met. Evidence: the catalog row selected, bounded context metadata when used, and the procedure's own output evidence. Output: one routing announcement followed by the selected workflow.

No-context invocation

When /hi or $hi has no following text, do not load the catalog. Read protocols/hi-menu.md, ask its two-stage menu, then load only the chosen procedure. Its recommendation branch may dispatch Librarian.

Contextual routing

Run uv run scripts/intent_coverage.py catalog --examples and read it: one line per harness/intents.toml row (plus any private rows from the local overlay, marked (private)) with its description, example phrases, and dispatch tag. Pick the row whose description fits the request's primary intent; examples illustrate a row, they are not triggers. Judge the whole message, not a keyword inside it: a URL inside a day's narrative is capture, not reading; a question that cites a link is a question. A date-prefixed factual narrative with no analytical question is capture. Requests that fit no row are general, which hands off to the runtime's ordinary skill, app, agent, or tool routing. Never choose reflection because nothing else fit.

Clarify only when two rows fit equally well and the choice changes what runs, or when a low-confidence route would open a file, call an external service, write, or start a multi-agent chain. Offer the plausible rows plus general. Do not clarify a request whose reading is obvious in context.

Log every contextual route once, best-effort, before dispatch:

uv run scripts/intent_coverage.py intent-log --quiet \
  --input "<raw hi text>" --runtime <claude-code|codex> \
  --match-kind <routed|general|clarified> --intent <name> \
  [--candidates a,b --clarified-to <name>] [--final-dispatch <label>]

routed is a confident pick, general a handoff (add --final-dispatch <capability> once you know what ran), clarified a route the user chose from candidates. If the user redirects a confident route after the announcement, log a second line before continuing: --match-kind corrected --intent <announced> --clarified-to <actual>. That line is the only record of a false hit. The ledger is the coverage signal for protocols/intent-coverage.md; a failed write never blocks dispatch.

Load and dispatch

Announce Routing as intents.<name> → <agents>, adding (parallel) when the catalog tag says so. For general, say No Atelier intent matched; handing off semantically → <capability> and execute intents.general.

When the row declares profile_reads, run:

uv run scripts/context_bundle.py --intent <name>

The helper copies only the route-selected profile and continuity sources into an isolated staging tree, then emits a standard Repomix XML artifact. It applies the row's hard context_budget_tokens ceiling without truncation. Reuse the artifact; do not reload the same sources. Skip it when profile_reads is empty unless the procedure requests a specific source.

Read only the row's procedure file and execute it as the selected workflow. When the row is parallel, dispatch the declared initial agents in one native batch. Procedure-level parallel steps follow the same rule. All writes retain the approvals and daily-note boundaries from AGENTS.md.

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.