usage-cost-plugin/catalog.json at master · Atishyy27/usage-cost-plugin · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
{
  "$comment": "Fork me. Each rule maps a usage signal to a plugin/skill recommendation. Conditions reference dot-paths into the fingerprint (see scripts/core.js -> fingerprint()). 'all' = every condition must hold; 'any' = at least one. op is one of > >= < <= == . Lower priority number = shown first.",
  "rules": [
    {
      "id": "context-compression",
      "priority": 1,
      "all": [{ "path": "counts.cacheReadTokens", "op": ">", "value": 5000000 }],
      "title": "Your sessions carry heavy context",
      "recommend": "headroom (context compression) + fresh sessions",
      "why": "Millions of cache-read tokens means you re-process a large context every turn. That is usually the single biggest slice of spend and latency.",
      "how": "Use the headroom plugin's compress on big tool outputs, and start a fresh session (/clear) for unrelated tasks so you stop dragging a huge context tail."
    },
    {
      "id": "ship-flow",
      "priority": 2,
      "all": [{ "path": "counts.gitOps", "op": ">=", "value": 8 }],
      "title": "You run a lot of git/gh by hand",
      "recommend": "compound-engineering:ce-commit-push-pr",
      "why": "Frequent manual git/gh calls are exactly what a commit -> push -> PR skill automates, with value-communicating messages.",
      "how": "Invoke /ce-commit-push-pr (compound-engineering plugin) to commit, push, and open a PR in one step."
    },
    {
      "id": "review-before-ship",
      "priority": 2,
      "all": [{ "path": "counts.editWrite", "op": ">=", "value": 25 }],
      "title": "Lots of edits, no review pass",
      "recommend": "/code-review and /simplify",
      "why": "A high edit/write volume is where bugs and cruft accumulate. A focused review + simplify pass on the diff catches them before they ship.",
      "how": "Run /code-review for correctness and /simplify for reuse/clarity on your working diff."
    },
    {
      "id": "research-depth",
      "priority": 3,
      "any": [
        { "path": "counts.webFetches", "op": ">=", "value": 8 },
        { "path": "counts.webSearches", "op": ">=", "value": 4 }
      ],
      "title": "You do a fair amount of web research",
      "recommend": "deep-research skill",
      "why": "Repeated ad-hoc web search/fetch is better served by a harness that fans out, fetches, verifies, and synthesizes a cited report.",
      "how": "Run /deep-research with your question for a multi-source, fact-checked writeup."
    },
    {
      "id": "permission-fatigue",
      "priority": 3,
      "all": [{ "path": "counts.bashCalls", "op": ">=", "value": 30 }],
      "title": "Heavy shell use = lots of approval prompts",
      "recommend": "fewer-permission-prompts skill",
      "why": "A high volume of Bash calls means repeated permission prompts for commands you already trust. An allowlist removes that friction.",
      "how": "Run /fewer-permission-prompts to scan your transcripts and add a safe allowlist to .claude/settings.json."
    },
    {
      "id": "test-runner",
      "priority": 3,
      "all": [{ "path": "counts.testOps", "op": ">=", "value": 5 }],
      "title": "You run test suites often",
      "recommend": "test-runner agent + /verify",
      "why": "Delegating test runs to a dedicated agent (that reads output and reports root cause) is faster than driving them by hand each time.",
      "how": "Ask for the test-runner agent to run a targeted subset, or /verify to exercise a change end-to-end."
    },
    {
      "id": "use-subagents",
      "priority": 4,
      "all": [
        { "path": "meta.totalToolCalls", "op": ">=", "value": 100 },
        { "path": "counts.subagents", "op": "==", "value": 0 }
      ],
      "title": "High activity, but no parallel subagents",
      "recommend": "Explore / general-purpose subagents (Task)",
      "why": "Fanning out independent reads/searches to cheap subagents (e.g. Haiku) is faster and far cheaper than doing them serially in the main thread.",
      "how": "For parallel or independent work, spawn subagents; keep the main, expensive model for the hard reasoning."
    }
  ]
}