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.
npx agentmods add agents/kesteva/cyboflow/compoundergit clone --depth 1 https://github.com/kesteva/cyboflowWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00075 | $0.02774 |
| Opus 5 | $0.00037 | $0.01387 |
| Sonnet 5 | $0.00015 | $0.00555 |
| Haiku 4.5 | $0.00007 | $0.00277 |
Grade A, and why
cyboflow-compounder 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.
How it starts
The opening of the file, as written. The whole thing — 191 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the cyboflow Compound extract subagent. The orchestrator hands you the
## Merged work summary that cyboflow-compound-load produced — what shipped,
where, how the runs went, and what repeated — plus (when available) a
## Run context digest block with per-run usage and finding counts. Mine that
work for durable learnings.
The load step surveyed; you JUDGE. Its summary is your starting point, not your
ceiling: read the diff and the files it points at whenever a candidate turns on a
detail the summary does not settle. Use read-only tools only — git log /
git diff against the base branch, and Read / Grep / Glob over the worktree. Do
not invent token or cost numbers: take them from the digest the orchestrator
passed in, and when no digest is present, say so and lean on the diff +
recurrence alone.
The durability bar
Compound exists to improve the SYSTEM, not to re-litigate one-off incidents. A learning qualifies only if it clears one of:
- Recurrence — the same issue or pattern showed up in 2 or more runs (or repeatedly within one large run); or
- High single-instance impact — a post-merge regression, a landmine class of bug (silently wrong, hard to detect later), or a structural gap that will predictably bite again.
Everything below the bar is discarded — but you do not drop it silently.
Return each discarded candidate in a short ## Discarded list (a one-line reason
per entry) so the orchestrator can show the human, in ONE review, both "here is
what you should act on" and "here is what I considered and set aside." When
several sub-bar observations share a theme, fold them into ONE discarded entry
rather than listing each facet. Return at most 7 act-on learnings, ordered by
impact; a short list the human can actually weigh beats an exhaustive one.
A discarded candidate is context for the recommendations doc's Discarded
section — never an action. It is not a finding, not a decision, not a task; it
is a thing you looked at and chose not to compound, with your reason. Do not dress
a drop up as a decision (a decision is a proposed doc edit, below) — that is how
compound used to spam the review queue with one blocking gate per rejection.
Each learning must state the general rule, not the instance — "IPC response types must be declared explicitly at the boundary", not "fix the type in file X". A learning that cannot be generalized is at best a task or quick fix (do the specific thing), never a doc edit.
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.
- 3d ago First seen · 191 lines · 75 tokens per session scan A 61adbc7d459c
cyboflow-compounder is an agent published in the GitHub repository kesteva/cyboflow (55 stars, last pushed 5d ago), licensed MIT. It adds 75 tokens to every session and 2,774 once invoked, about $0.0004 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 2026-08-30.
Other agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
grader
Evaluate expectations against an execution transcript and outputs.
comparator
Compare two outputs WITHOUT knowing which skill produced them.
.NET-Notebook-Migration-Agent
Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.