ecc — everything-claude-code

A plugin pack from an Anthropic hackathon winner: a large catalog of commands, agents, skills, and hooks, invoked here through two commands — /plan and /build-fix.

Upstream

Performance

Metric Value
Combined rank 2 / 9
Combined z̄ +0.244
Feature z (rank) +0.091
Bugfix z (rank) +0.680
Refactor z (rank) −0.039

The feature/refactor results are middle-of-pack; the run is pulled into a top-tie almost entirely by a strong bugfix trial.

Mechanism

ECC ships a sprawling plugin surface — its own documentation claims 38 agents, 156 skills, and 72 legacy command shims at v1.10.0, with additional components for cost, security, and operator workflows. The config directory confirms an MIT plugin registered through the Anthropic marketplace and a single enabledPlugins entry (everything-claude-code@everything-claude-code); no skills were manually re-keyed into ~/.claude/.

For the benchmark, three entry points out of the full catalog were reached:

How this benchmark invoked it

scripts/manual-bench.sh selects a different command per task:

That split is load-bearing. plan paths begin with a planner subagent round trip before any file edits; build-fix goes straight to investigation-and-patch without a separate plan gate.

What actually happened in transcripts

Feature (Mode 2 CD Batch, +0.091 z). Two sessions: a scoped planner pass, then a ~263-turn implementation. The plan subagent emitted a structured requirements/risks/steps document; the user replied yes; the main agent then read ~15 files, edited the Mode 1 strategy and interface, and wrote a new td-cd-mode2.strategy.ts plus cd-aging.util.ts and its spec. 11 files changed (+477 / −11). Final jest runs show 18/18 new Mode-2 tests and 70/70 td-cd-savings tests green; 75 tests pass, 0 fail; 14 eslint errors and 7 warnings remained at commit time. The agent correctly dismissed pre-existing failures in an unrelated ntd-cd-savings suite rather than chasing them.

Bugfix (SHP-2376 near-maturity batch, +0.680 z). Single session. /build-fix read TASK.md, dispatched two Explore subagents in parallel to map savings-cd and the core CD service, and narrowed to the right file (savings-cd-batch-data-source.ts) on the first attempt. The patch added a maturity filter where the bug report said it was missing, and a new spec file reproduced the bug. Before each edit the agent ran a “fact-forcing gate” listing importers, affected classes, and data impact — a pattern that comes from the plugin’s agent prompts, not from the task brief. Test result: savings-cd suite 62/62 green including 5 new tests. auto-metrics.json records 89 passed / 15 failed overall; the failures are in unrelated suites and were explicitly attributed by the agent to pre-existing mock issues. 2 files changed (+177 / −1).

Refactor (SHP-2317, −0.039 z). 1009-line session. After the planner pass, the agent executed 10 edits across libs/core, moved cdBatchId from Scheme to TSSchemeSetting, introduced a CDBatchInfo projection DTO to replace CDBatch in strategy signatures, and rewired td-cd-savings.service.ts plus both mode2.strategy and savings.service specs. The same fact-forcing preamble ran before each edit. 10 files, +107 / −52. All tests green, 6 eslint errors / 7 warnings. Result is clean but low-reward: the refactor landed below the field mean.

Why it ranked where it did

The bugfix score carries the run. The /build-fix command matched the task shape exactly: a QA report with a named file to fix, explicit test commands, and a hard scope boundary. The plugin’s fact-forcing gate and parallel Explore dispatch gave tight first-cut scoping and a reproduction-test that other tools in this benchmark did not consistently produce.

Feature and refactor used the same /plan entry point and got only moderate lift. The planner subagent adds a clean upfront structure but does not keep pushing during implementation; the main thread then does ordinary Claude Code work and drifts into the same lint/style trade-offs that put the run behind pure and bmad on those tasks.

Strengths and failure modes

Strengths observed: crisp task-to-command mapping when the harness supplies it (plan vs build-fix); planner subagent produces reviewable plans with explicit confirm gates; fact-forcing gate catches blast-radius before each edit; willingness to separate pre-existing failures from regressions.

Failure modes observed: 14 eslint errors on the feature diff, 6 on the refactor — the workflow does not round-trip lint before committing. Large catalog is mostly dormant (TDD, quality-gate, code-review commands never fire unless the harness picks them). On tasks without a clear “this is a bug” framing (refactor), the tool reduces to vanilla Claude with an extra planning prelude.

Unverified items: 140K-star / 21K-fork counts on the upstream README — README claim only, not checked against GitHub.

References

Observed in trial timelines

Across all 8 trials, ecc dispatches the planner subagent (everything-claude-code:planner on feature/refactor, Explore pair under /build-fix on bugfix) — mean 2.0 on feature, 1.5 on bugfix, 1.0 on refactor. No skill files appear as Read events; ecc’s planner is loaded as a subagent prompt, not a plugin file.

Detail: see the per-trial timeline files linked below.

Trial timelines

Per-trial event timelines auto-extracted from session-logs/*.jsonl — skill activations, plugin/skill file reads, subagents dispatched, code mutations, Bash usage:

Trial timelines

Per-trial session execution extracted from each trial's session-logs/*.jsonl. Each card shows the subagents dispatched, skill activations, Bash command mix, and the final diff. Switch task tabs to compare behaviour across feature, bugfix, and refactor trials.

t1 13:34 → 14:43 UTC · 68 min
3 commits11 files+477

“<command-message>everything-claude-code:plan</command-message> <command-name>/everything-claude-code:plan</command-name> <command-args>Read the PRD at docs/infina-product-docs/docs/core-products/td-cd/user-logic/[PRD] [T…”

  • Agents2
  • New files4
  • Edits16
  • Bash22
  • Sessions2
Bash command mix · 22 calls
  • tests 13
  • other 4
  • git ops 4
  • inspection 1
Subagents dispatched (2)
  • everything-claude-code:planner · Plan Mode 2 CD Batch implementation at 13:34
  • everything-claude-code:planner · Plan Mode 2 CD Batch implementation at 14:23
Subagent transcripts (2)
  • agent-a06a5f40777a… — I need to plan the implementation of Mode 2 CD Batch for the TD-CD product. **Task**: Read the PRD a… [Read×28, Grep×12, Glob×2]
  • agent-a4deb7d17b7e… — Read the PRD at docs/infina-product-docs/docs/core-products/td-cd/user-logic/[PRD] [TD-CD] User stor… [Read×25, Grep×11, Glob×3]
New files created (4)
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.spec.ts
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.ts
  • libs/core/src/domain/utils/cd-aging.util.spec.ts
  • libs/core/src/domain/utils/cd-aging.util.ts
t2 14:37 → 15:19 UTC · 42 min
2 commits7 files+454

“<command-message>everything-claude-code:plan</command-message> <command-name>/everything-claude-code:plan</command-name> <command-args>Read the PRD at docs/infina-product-docs/docs/core-products/td-cd/user-logic/[PRD] [T…”

  • Agents1
  • New files6
  • Edits4
  • Bash32
  • Todos7
Bash command mix · 32 calls
  • tests 15
  • other 10
  • inspection 5
  • lint/format 1
  • git ops 1
Subagents dispatched (1)
  • everything-claude-code:planner · Plan Mode 2 CD Batch implementation at 14:37
Subagent transcripts (1)
  • agent-a70162e32ba7… — Create a comprehensive implementation plan for adding Mode 2 CD Batch to the TD-CD product in this N… [Grep×13, Read×10, Glob×3]
New files created (6)
  • libs/core/src/domain/savings-cd/index.ts
  • libs/core/src/domain/savings-cd/td-cd-aging-days.util.spec.ts
  • libs/core/src/domain/savings-cd/td-cd-aging-days.util.ts
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.spec.ts
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.ts
  • libs/core/src/port/service/td-cd-mode-strategy.port.ts
t3 04:36 → 04:57 UTC · 21 min
2 commits8 files+410

“Read the PRD at docs/infina-product-docs/docs/core-products/td-cd/user-logic/[PRD] [TD-CD] User stories - Mode 2 CD Batch.md and study the existing Mode 1 implementation in libs/core/src/domain/savings-cd/ and libs/savin…”

  • Agents3
  • New files4
  • Edits8
  • Bash25
Bash command mix · 25 calls
  • tests 17
  • typecheck 3
  • git ops 3
  • inspection 1
  • other 1
Subagents dispatched (3)
  • Explore · Read PRD document at 04:36
  • Explore · Study Mode 1 implementation patterns at 04:36
  • Explore · Find TD-CD entities and enums at 04:37
Subagent transcripts (3)
  • agent-a17e8457f0d1… — Read the full contents of this file and report back everything in it: docs/infina-product-docs/docs/… [Read×1]
  • agent-a5b37892a600… — Search the entire codebase for TD-CD related entities, enums, constants, and types. Specifically fin… [Read×21, Bash×12, Grep×8, Glob×1]
  • agent-abc1a6a849a0… — Thoroughly explore the existing Mode 1 TD-CD implementation to understand the patterns, file layout,… [Bash×30, Read×26]
New files created (3)
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.spec.ts
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.ts
  • libs/core/src/port/service/td-cd-mode-strategy.port.ts
t4 09:05 → 09:34 UTC · 29 min
2 commits6 files+408

“<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you t…”

  • Agents2
  • New files8
  • Edits15
  • Bash21
  • Sessions3
Bash command mix · 21 calls
  • tests 10
  • other 7
  • git ops 4
Subagents dispatched (2)
  • Explore · Read PRD and Mode 1 implementation at 09:14
  • Explore · Explore Mode 2 dependencies at 09:16
Subagent transcripts (2)
  • agent-a757d3199216… — I need to understand the Mode 2 CD Batch feature for a TD-CD product. Please do the following thorou… [Read×22, Glob×7, Grep×4, Bash×3]
  • agent-a8c9f654f123… — I need to understand several key implementation details for building a Mode 2 strategy. Please find… [Read×15, Bash×13, Glob×9, Grep×5]
New files created (4)
  • libs/core/src/domain/savings-cd/cd-aging-days.util.spec.ts
  • libs/core/src/domain/savings-cd/cd-aging-days.util.ts
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.spec.ts
  • libs/core/src/domain/savings-cd/td-cd-mode2.strategy.ts
t1 16:41 → 16:48 UTC · 7 min
2 commits2 files+177

“<command-message>everything-claude-code:build-fix</command-message> <command-name>/everything-claude-code:build-fix</command-name> <command-args>A QA bug report was filed. Read docs/benchmark/TASK.md for the full report:…”

  • Agents2
  • New files2
  • Edits3
  • Bash13
Bash command mix · 13 calls
  • tests 5
  • other 5
  • git ops 2
  • inspection 1
Subagents dispatched (2)
  • Explore · Explore savings-cd codebase at 16:41
  • Explore · Explore core service for CD at 16:41
Subagent transcripts (2)
  • agent-a177d0c06d36… — Explore the savings-cd library in /Users/randytran/Codes/ai-tool-benchmark/runs/shp2376/ecc-t1/libs/… [Read×24, Bash×15, Glob×2, Grep×2]
  • agent-aa162bc39343… — Explore the core app's savings CD service at /Users/randytran/Codes/ai-tool-benchmark/runs/shp2376/e… [Bash×10, Read×9, Grep×5, Glob×2]
New files created (1)
  • ecc-t1/libs/savings-cd/src/domain/savings-cd-batch-data-source.spec.ts
t2 15:59 → 16:16 UTC · 16 min
2 commits2 files+152

“<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you t…”

  • Agents1
  • New files2
  • Edits3
  • Bash16
  • Sessions2
Bash command mix · 16 calls
  • tests 8
  • other 5
  • git ops 2
  • inspection 1
Subagents dispatched (1)
  • Explore · Explore Savings CD codebase at 16:07
Subagent transcripts (1)
  • agent-aecfedf0d4f1… — I'm investigating bug SHP-2376 in a NestJS monorepo at /Users/randytran/Codes/ai-tool-benchmark/runs… [Read×16, Grep×9, Bash×7, Glob×5]
New files created (1)
  • ecc-t2/libs/savings-cd/src/domain/savings-cd-batch-data-source.spec.ts
t1 03:57 → 04:13 UTC · 16 min
2 commits10 files+107

“<command-message>everything-claude-code:plan</command-message> <command-name>/everything-claude-code:plan</command-name> <command-args>A refactor has been requested by the tech lead. Read docs/benchmark/TASK.md for the f…”

  • Agents1
  • Edits31
  • Bash11
Bash command mix · 11 calls
  • other 5
  • git ops 3
  • tests 2
  • inspection 1
Subagents dispatched (1)
  • Explore · Explore codebase for refactor at 03:57
Subagent transcripts (1)
  • agent-a7945124917c… — Explore the codebase at /Users/randytran/Codes/ai-tool-benchmark/runs/shp2317/ecc-t1 with thoroughne… [Bash×32, Read×22, Glob×5]
t2 06:34 → 06:47 UTC · 13 min
2 commits9 files+95

“<command-message>everything-claude-code:plan</command-message> <command-name>/everything-claude-code:plan</command-name> <command-args>A refactor has been requested by the tech lead. Read docs/benchmark/TASK.md for the f…”

  • Agents1
  • Edits28
  • Bash13
Bash command mix · 13 calls
  • other 6
  • git ops 3
  • tests 2
  • inspection 2
Subagents dispatched (1)
  • Explore · Explore current codebase structure at 06:34
Subagent transcripts (1)
  • agent-add6080a04ac… — I need to understand the current state of a refactor task (SHP-2317) in the codebase at /Users/randy… [Read×16, Glob×12, Grep×5, Bash×2]