cascade-runner

A helper agent that coordinates PDF searching for a batch of research references. It delegates each download to the PDF-acquisition tool and combines the results.

In plain words
What is it for?
Use it to process named references or all references in a chosen state, with an optional limit, and report completed, pending, blocked, or withdrawn items.
Why use it?
It keeps large batches of references organised and leaves the main agent with a concise list of successes and references needing decisions.

Agent

Part of the paper-trail plugin — 6 skills, 22 commands, 6 agents, 3 hooks shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/roomi-fields/paper-trail/cascade-runner
Clone the repo
git clone --depth 1 https://github.com/roomi-fields/paper-trail

Or install paper-trail, the plugin that ships this one along with the rest of its 6 skills, 22 commands, 6 agents, 3 hooks.

Per session 60 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 641 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00060 $0.00641
Opus 5 $0.00030 $0.00320
Sonnet 5 $0.00012 $0.00128
Haiku 4.5 $0.00006 $0.00064

Measured 3d ago against content hash f51b3eed3497, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

cascade-runner scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/cascade-runner.md · 78 lines

How it starts

The opening of the file, as written. The whole thing — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Sub-agent : cascade-runner

Role

Orchestrate the worker B cascade across a batch of references. Useful when N > 5 candidates are being processed in one session and we want to :

  • Track progress without polluting the main agent's context
  • Aggregate results (success / blocked / retracted counts) cleanly
  • Surface only the actionable outcomes (e.g., refs that need human decision)

For single-ref acquisitions, the pdf-cascade skill suffices.

Input contract

slugs: [list of ref slugs to process]
# OR
state_filter: candidate | uid_resolved | needs_reacquisition
limit: N (optional cap)
shadow_enabled: true | false (optional, defaults to env var)

What this agent does

  1. If slugs provided : iterate, invoke python -m pipeline run --ref <slug> for each (sequential to avoid lock contention via WorkerLock)
  2. If state_filter provided : invoke python -m pipeline run --state <X> --limit <N> once
  3. Parse the worker B output (recap session line) for each invocation
  4. Aggregate :
    • success_slugs[] : refs reaching page1_validated
    • pending_slugs[] : refs reaching awaiting_rtfm_ocr (OCR queued)
    • blocked_slugs[] : refs reaching blocked_human:* (with reason)
    • retracted_slugs[] : refs reaching retracted
  5. Return structured summary to caller

Output contract

{
  "batch_size": N,
  "success_slugs": ["arnold_1982", "smith_2020", ...],
  "pending_slugs": ["lerdahl_2001", ...],
  "blocked_slugs": [
    {"slug": "chemillier_2003", "reason": "title_mismatch"},
    ...
  ],
  "retracted_slugs": [],
  "elapsed_seconds": 145.2,
  "errors": []
}

Constraints

  • Sequential invocation (worker B has a WorkerLock preventing concurrent cmd_run sessions)
  • No retries on transient failures (the worker B's circuit-breakers and cascade exhaustion handle that)
  • Never modifies registry directly — only via worker B CLI

When NOT to invoke

  • Single ref by slug : use pdf-cascade skill directly (no need for sub-agent overhead)
  • Semantic decisions (curator role) : use sota-auditor skill
  • Doctor / invariant checks : use registry-doctor skill

Read the full file on GitHub · 78 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 78 lines · 60 tokens per session scan A f51b3eed3497

Subscribe to this mod's changes

cascade-runner is an agent published in the GitHub repository roomi-fields/paper-trail (5 stars, last pushed 10d ago), licensed MIT. It adds 60 tokens to every session and 641 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.