agentsclimarketplace

Threnody fast review

Skill timjensgrossinger/threnody/skills/threnody-fast-review

Fast broad code review swarm: one read-only host agent per file, then one synthesis agent. Use when asked for a fast swarm review, one-agent-per-file review, broad review sweep, ordinary swarm review, or when reviewing many files where coverage and speed matter more than per-dimension depth.From its SKILL.md

Install
npx -y skills add timjensgrossinger/threnody --skill threnody-fast-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.4 KB, as published. Nobody here has run it

Threnody fast review

Purpose

Use this when a review spans many files and should not collapse into a tiny number of agents. This mode trades the deeper file × dimension review swarm for faster file-level parallelism:

wave 1: one read-only review agent per file
wave 2: one synthesis agent

This is the default for broad swarm review. Use threnody-swarm-review only when the user explicitly asks for deep review, security-critical audit, threat-modeling, or a named specialist dimension.

Tiering is medium by default. Ordinary risk words such as auth, token, or secret add security attention inside the file reviewer but do not automatically force high tier. High tier is reserved for explicit deep/security-critical wording, concrete exploit primitives, or large/dense files.

The global swarm.max_agents cap still applies. If the requested count is clamped, report the requested_vs_effective_agent_count field and tell the operator which files were not reviewed.

Workflow

0. Resolve files

Use file paths from the user's message. If none are provided, resolve changed source files:

git diff --name-only HEAD
git diff --name-only --cached

Filter out generated files, lock files, binary assets, Markdown docs, and large schemas unless the user explicitly asks to review them.

1. Run cheap deterministic checks

Run relevant linters or type checks first when available. Keep failures as context for synthesis; do not stop the swarm solely because a linter found findings.

2. Call execute_swarm

Use the FAST_REVIEW: sentinel and request one agent per file plus one synthesis agent:

execute_swarm(
  task="FAST_REVIEW: <space-separated file paths>",
  topology="dag",
  max_agents=<number_of_files + 1>,
)

The sentinel activates fast file-level review in the heuristic planner.

3. Execute host_spawn_waves

On awaiting_host_execution: true + host_spawn_waves:

  • Spawn every wave-1 agent in parallel.
  • Pass each agent its prompt, model, tier, subagent_type, and target file.
  • Keep review read-only. Never Write or Edit reviewed files.
  • Run the synthesis wave after all file agents finish.

4. Report

Use report_host_swarm_complete once at terminal in batch mode, or report_host_wave per wave in inline mode, following learning_report_contract.

Before final reporting, run a cheap targeted verifier only for synthesized HIGH or CRITICAL findings. Do not run a second full review swarm; verify the specific finding with file:line grounding and mark it valid, false_positive, or needs_more_evidence.

Must

  • Use FAST_REVIEW: exactly.
  • Set max_agents to number_of_files + 1 unless the user asks for a lower cap.
  • Report clamping if effective agents are lower than requested agents.
  • Preserve read-only behavior.

Do not

  • Use this for a deep security-critical audit where per-dimension specialists are needed. Use threnody-swarm-review instead.
  • Collapse a large review into one general agent unless the user explicitly asks.
  • Escalate every file to high tier just because it mentions auth, token, or credentials.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.