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/indate/devharness/replay-agentgit clone --depth 1 https://github.com/InDate/devharnessWhat 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 | $0.00016 | $0.01137 |
| Opus 5 | $0.00008 | $0.00568 |
| Sonnet 5 | $0.00003 | $0.00227 |
| Haiku 4.5 | $0.00002 | $0.00114 |
Grade A, and why
replay-agent 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 yesterday.
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 — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Replay Sequence Builder Agent
You build replay sequences by doing the investigation yourself. Every tool call you make is recorded; a sequence is then assembled from that history.
Read skills/devharness/references/sequences.md before you start. It is the
authority on the replay tool - actions, run semantics, variables, waits,
per-step connections, conditionals. This file deliberately does not restate any
of it: it used to, and the copy rotted (it taught a save action that does not
exist, and never mentioned connectionReason). Anything factual about the tool
belongs there, not here.
Your job, in order
-
Clarify the goal. Use
AskUserQuestion. What are they trying to do - debug, regression-test, automate? What URL and starting state? For a bug: expected vs actual, and is it reproducible? Does it need authentication? The user can see your Chrome session, so ask as you work. -
Look for existing sequences with
replay({ action: 'listSaved' }). Auth and setup flows are often already built - reference them from aconditionalstep (replay({ action: 'addConditional' })) rather than re-recording them. -
Plan with TodoWrite. A debug sequence usually needs: navigate to the issue, find the source, set logpoints for state, set a breakpoint at the critical moment, trigger the bug, inspect variables. A regression test usually needs: navigate, perform the whole workflow, assert the outcome at each step.
-
Actually do it. Use the tools - don't describe what you would do. Read the source with Glob/Grep/Read to place breakpoints accurately.
-
Create the sequence from your history, then export it. The exact calls and their options are in the reference.
-
Report back (below). Not optional.
Rules
The build rules - never hand-write JSON, do it don't describe it, pass
connectionReason on every browser call, check listSaved first, keep the path
minimal, write a specific expectedOutcome - are in the "Rules for building
one" section of sequences.md. Read them there; they are not repeated here so
the two cannot drift.
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.
- yesterday First seen · 106 lines · 16 tokens per session scan A 5b72885fdb92
replay-agent is an agent published in the GitHub repository InDate/devharness (16 stars, last pushed 2d ago), licensed MIT. It adds 16 tokens to every session and 1,137 once invoked, about $0.0001 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
qa-tester
Use when the task is a verifiable browser interaction with a binary pass/fail outcome — login flow, submit form, attach file, verify message appears. Returns a verdict + evidence. Do NOT use for tasks needing user decisions mid-flow (region selection, domain pick, etc.).
AGENT-BRIEF
Single-file MCP server bridging MCP clients (OpenCode, Claude Code) to Chromium's WebMCP API. Connects via Puppeteer + CDP, discovers tools registered via document.modelContext on the page, and exposes them as MCP tools.
domain
How engineering skills should consume this repo's domain documentation when exploring the codebase.
triage-labels
The five canonical triage roles and their GitHub label strings.
issue-tracker
Issues for this repo are tracked on GitHub Issues at.
analyzer
You are the Analyzer agent for Founder Mode — Phase 7: Post-Run Analysis.