style: remove all em dashes from skill files

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-10-29 18:05:20 +08:00 · 2026-07-09 18:22:00 -07:00
parent ff5023c130
commit e3ee9fbaa7
5 changed files with 25 additions and 25 deletions
@@ -2,16 +2,16 @@

 **One model is a bottleneck. A team with one brain, twenty hands, and a board advisor is not.**

-This skill turns your coding agent into the orchestrator of a three-tier model team. Big tasks get split into self-contained briefs, blasted across cheap parallel workers, verified one by one, and judged by a stronger model exactly twice — before the work starts and before it ships.
+This skill turns your coding agent into the orchestrator of a three-tier model team. Big tasks get split into self-contained briefs, blasted across cheap parallel workers, verified one by one, and judged by a stronger model exactly twice: before the work starts and before it ships.

 <img width="3004" height="1408" alt="advisor_skill" src="https://github.com/user-attachments/assets/6f5dc5e8-6828-4598-b23c-72ede97fa238" />

 ## The team

-| Role | Default model <sub>(July 2026 — swap freely)</sub> | What it does | What it never does |
+| Role | Default model <sub>(July 2026, swap freely)</sub> | What it does | What it never does |
 |---|---|---|---|
 | **Orchestrator** | GPT-5.6 | Frames success criteria, plans waves, dispatches briefs, verifies every result, synthesizes the deliverable | Worker-level grunt work |
-| **Workers** | Gemini 3.5 Flash | One self-contained subtask each, in parallel, stateless — each sees only its brief, with tool access when the subtask needs it | Talk to each other, expand scope, get a second chance on the same call |
+| **Workers** | Gemini 3.5 Flash | One self-contained subtask each, in parallel, stateless; each sees only its brief, with tool access when the subtask needs it | Talk to each other, expand scope, get a second chance on the same call |
 | **Advisor** | Claude Fable 5 | Plan review before any dispatch, taste pass before delivery, called mid-run only at commitment boundaries | Execute anything |

 The economics are the point: cheap parallel generation where volume wins, expensive judgment only where it changes a decision. Every run states a budget up front, sized to the plan, and never spends past it silently: running out means an honest report or an explicit ask, not quiet burn. Models are knobs: the tier pattern is the durable part, the defaults were current in July 2026.
@@ -20,9 +20,9 @@ The economics are the point: cheap parallel generation where volume wins, expens

 Multi-model loops usually die from context leaks, silent partial failures, or judgment applied too late. Each has a rule here:

- **Stateless briefs** ([references/worker-brief.md](references/worker-brief.md)) — every dispatch carries its full inputs and acceptance criteria inline. No shared context, no "as discussed above." Briefs travel as temp files — handed to `agy` as a quoted expansion, or jq-built into API payloads on the fallback path — never interpolated into shell strings.
- **Verify before merge** — every result is judged against its own acceptance criteria: PASS, FIX (redispatched with the named failure), or ESCALATE. No silent partial passes, no hand-patching.
- **The advisor is a critic, not an executor** ([references/advisor-consult.md](references/advisor-consult.md)) — verdict, ranked risks, concrete fixes, under 300 words. Every note gets applied or explicitly rebutted.
+- **Stateless briefs** ([references/worker-brief.md](references/worker-brief.md)): every dispatch carries its full inputs and acceptance criteria inline. No shared context, no "as discussed above." Briefs travel as temp files (handed to `agy` as a quoted expansion, or jq-built into API payloads on the fallback path), never interpolated into shell strings.
+- **Verify before merge**: every result is judged against its own acceptance criteria: PASS, FIX (redispatched with the named failure), or ESCALATE. No silent partial passes, no hand-patching.
+- **The advisor is a critic, not an executor** ([references/advisor-consult.md](references/advisor-consult.md)): verdict, ranked risks, concrete fixes, under 300 words. Every note gets applied or explicitly rebutted.

 ## Install

@@ -36,7 +36,7 @@ Or copy this folder into your agent's skills dir (`~/.claude/skills/`, `~/.codex

 ## Use it

-> "This is too big for one pass — orchestrate it across a model team."
+> "This is too big for one pass. Orchestrate it across a model team."
 > "Fan this out: research all 12 competitors in parallel and synthesize."
 > "Run the advisor-worker loop on this."

@@ -53,6 +53,6 @@ advisor-orchestrator-worker/
 └── references/fallbacks.md           # the Gemini and Anthropic API fallback calls
 ```

-Evals live repo-side in `agent_skills/evals/advisor-orchestrator-worker/` — you install only what runs.
+Evals live repo-side in `agent_skills/evals/advisor-orchestrator-worker/`; you install only what runs.

 Part of [awesome-llm-apps](https://github.com/Shubhamsaboo/awesome-llm-apps) · Apache-2.0 · Last verified: July 2026
@@ -33,9 +33,9 @@ on another shell, run them with `bash -c`.

 ## The team

- **Workers (default: Gemini 3.5 Flash via the Antigravity CLI, `agy`)** —
-  stateless generation units, with tools (web search, file work) when a
-  subtask needs them. Never interpolate a brief into a shell string —
+- **Workers (default: Gemini 3.5 Flash via the Antigravity CLI, `agy`)**: stateless
+  generation units, with tools (web search, file work) when a
+  subtask needs them. Never interpolate a brief into a shell string;
  briefs carry quotes and arbitrary text, so that is a shell-injection
  bug. Write each brief to a temp file and dispatch each worker from
  its own EMPTY temp dir (no `.antigravity.md` or project context
@@ -57,18 +57,18 @@ on another shell, run them with `bash -c`.
  Chunk every wave into batches of 3 (Antigravity quota is shared
  across its app, CLI, and SDK). Start each batch with `pids=()`, reap
  each worker with its own `wait "$pid"` (a collective wait reports
-  only the last status), and read each `$out` in dispatch order — a
-  shared stdout hands verify interleaved output. Non-zero exit or an
+  only the last status), and read each `$out` in dispatch order,
+  since a shared stdout hands verify interleaved output. Non-zero exit or an
  empty `$out` is a failed dispatch: retry it through the Gemini API
  fallback in `references/fallbacks.md` when a key is set (no key:
  ESCALATE), and record the switch on the status board. That fallback also takes over when
  agy is missing, and carries any brief too large (over ~100 KB) or
  too untrusted for a CLI argument (`agy -p` has no prompt-file
-  input). API workers run uncapped in parallel but have no tools — a
+  input). API workers run uncapped in parallel but have no tools, so a
  subtask that needs tools goes through agy or gets ESCALATE. Clean up
  all temp files at run end.

- **Advisor (default: Claude Fable 5 via the claude CLI)** — consult
+- **Advisor (default: Claude Fable 5 via the claude CLI)**: consult
  written to a temp file, passed on stdin (never inline in the
  command), behind a timeout so a hung consult can't stall the loop
  (perl's alarm; timeout(1) is missing on stock macOS):
@@ -100,13 +100,13 @@ on another shell, run them with `bash -c`.
 4. **Delegate.** Dispatch each wave using the format in
   `references/worker-brief.md`. Parallel background calls, then wait.
 5. **Verify.** Check every result against its own acceptance criteria,
-   and make the check exercise the deliverable itself — run the actual
+   and make the check exercise the deliverable itself: run the actual
   command, read the actual output. Grepping a README, testing
   something adjacent, printing True while exiting zero, or re-checking
   that a file exists proves nothing. Verdict per result: PASS, FIX
   (redispatch naming the specific failure), or ESCALATE. Never
   silently accept a partial pass; never hand-patch a substantive
-   failure — redispatch instead.
+   failure; redispatch instead.
 6. **Synthesize.** When all subtasks pass, assemble the deliverable.
   Resolve conflicts between worker outputs explicitly, never by
   averaging.
@@ -135,4 +135,4 @@ needs the user. Return: the deliverable, the plan, a verification
 ledger per subtask, advisor notes applied and rejected, and remaining
 risks. Print a one-line status board after each loop step: per subtask,
 its state (PENDING / DISPATCHED / PASS / FIX / ESCALATED), dispatch
-path, and retries — e.g. `W2: FIX → PASS | agy→api | 1 retry`.
+path, and retries, e.g. `W2: FIX → PASS | agy→api | 1 retry`.
@@ -7,7 +7,7 @@ Both fallbacks keep the primary paths' discipline: payloads built with
 failure, and a closing jq that unwraps the response envelope so the
 orchestrator reads plain text on either path.

-## Workers — bare Gemini API call
+## Workers: bare Gemini API call

 No tools. Run the full wave in parallel; the 3-batch cap is an agy
 quota rule and does not apply here.
@@ -24,7 +24,7 @@ api_key="${GEMINI_API_KEY:-$GOOGLE_API_KEY}"
 pids+=($!)
 ```

-## Advisor — bare Anthropic API call
+## Advisor: bare Anthropic API call

 Same model as the CLI path. The `fallbacks` parameter re-serves a
 safety-classifier refusal on Opus 4.8 inside the same call, so one
@@ -1,11 +1,11 @@
 # Worker Brief Format

-Every worker dispatch is one stateless call — an agy CLI run, or a
-Gemini API request on the fallback path — containing this brief.
+Every worker dispatch is one stateless call (an agy CLI run, or a
+Gemini API request on the fallback path) containing this brief.
 The worker has no memory, no follow-ups, and sees nothing but this text.
 Inputs must be pasted inline in full; never reference material the
 worker cannot see. For code subtasks that means the entrypoint, file
-layout, and exact run commands — or workers will invent their own.
+layout, and exact run commands; otherwise workers invent their own.

 ```
 You are a worker completing ONE subtask of a larger project. This brief
@@ -13,7 +13,7 @@
        "The final report includes a verification ledger and advisor notes applied or rebutted",
        "Parallel worker outputs are collected from per-worker scratch files (mktemp + redirect, wait, read in dispatch order), never from a shared interleaved stdout",
        "Workers dispatch via the agy CLI by default with --model pinned to gemini-3.5-flash (same model as the API fallback), waves chunked into batches of at most 3 with a wait between batches",
-        "Briefs and consults are written to temp files (handed to agy as a quoted expansion, jq-built payloads on the API fallback, consults passed on stdin) \u2014 never interpolated inline into shell strings",
+        "Briefs and consults are written to temp files (handed to agy as a quoted expansion, jq-built payloads on the API fallback, consults passed on stdin); never interpolated inline into shell strings",
        "The Gemini API fallback fails fast if neither GEMINI_API_KEY nor GOOGLE_API_KEY is set (no silent empty-key call)",
        "Each backgrounded worker runs in its own subshell so parallel cd/cleanup and traps never collide",
        "Worker and advisor calls carry a timeout so a hung call cannot stall the whole wave",
@@ -45,7 +45,7 @@
    },
    {
      "id": 4,
-      "prompt": "A worker's output failed its acceptance criteria \u2014 just fix it up yourself and move on, we're in a hurry.",
+      "prompt": "A worker's output failed its acceptance criteria; just fix it up yourself and move on, we're in a hurry.",
      "expected_output": "The orchestrator redispatches with the named failure instead of hand-patching a substantive failure, or escalates after a second failure; it explains why in one line without lecturing.",
      "expectations": [
        "Substantive failures are redispatched with the failed criterion quoted, not silently hand-patched",