tracer

tracer is an agent for Claude Code from hg-pyun/claude-code-marketplace. It costs 72 tokens per session (3,649 once invoked), scanned A, original, MIT.

An evidence-based debugging agent that traces an observed problem, such as an error response or wrong output, back to the responsible code path.

In plain words
What is it for?
Following stack traces, logs, HTTP responses, and incorrect results to likely files and lines, then handing the findings to an agent that can fix them.
Why use it?
It ranks competing explanations using supporting and conflicting evidence before someone changes the code.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: model in frontmatter.

Part of the dev-tools plugin — 11 skills, 1 command, 15 agents shipped together

Good fit Following stack traces, logs, HTTP responses, and incorrect results to likely files and lines, then handing the findings to an agent that can fix them.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/hg-pyun/claude-code-marketplace/tracer
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.

Clone the repo
git clone --depth 1 https://github.com/hg-pyun/claude-code-marketplace

Made for: Claude Code.

Or install dev-tools, the plugin that ships this one along with the rest of its 11 skills, 1 command, 15 agents.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for tracer

README.md
[![agentmods](https://agentmods.dev/badge/agents/hg-pyun/claude-code-marketplace/tracer.svg)](https://agentmods.dev/agents/hg-pyun/claude-code-marketplace/tracer)
Your own site
<a href="https://agentmods.dev/agents/hg-pyun/claude-code-marketplace/tracer"><img src="https://agentmods.dev/badge/agents/hg-pyun/claude-code-marketplace/tracer.svg" alt="Measured on agentmods" height="20"></a>
Per session 72 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,649 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 1 finding. A grade says what 26 rules found in the file — not that it is safe.
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 5.1 $0.00072 $0.03649
Opus 5 $0.00036 $0.01825
Sonnet 5 $0.00014 $0.00730
Haiku 4.5 $0.00007 $0.00365

Measured 7d ago against content hash 89a5dfe1bc2c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-07, from the pricing page.

Security

Grade A, and why

tracer scanned grade A with 1 finding 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 7d 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.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

You are NOT responsible for: fixing the code (delegate to `executor`); full architectural root-cause analysis or systemic RCA with recommendations (delegate to `debugger` or `architect`); running the request or collectin
plugins/dev-tools/agents/tracer.md · 232 lines

How it starts

The opening of the file, as written. The whole thing — 232 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are responsible for: reverse-tracing from observable effects (HTTP response, log line, stack trace, wrong output) to file:line code paths; enumerating competing hypotheses; scoring each hypothesis against collected evidence; and producing a ranked evidence chain for downstream agents.

You are NOT responsible for: fixing the code (delegate to executor); full architectural root-cause analysis or systemic RCA with recommendations (delegate to debugger or architect); running the request or collecting the live signal (the caller provides the effect; use curl-debug skill for live cURL execution); severity-rated diff review (delegate to reviewer).

<Use_When>

  • A caller has an observed effect (error response body, log line, stack trace, incorrect output field) and needs to know which code path produced it.
  • Competing explanations exist and need to be ranked by evidence before a fix is attempted.
  • curl-debug skill has captured a response and needs the reverse-trace portion formalized as a reusable agent step.
  • A downstream skill (e.g., ralph, autopilot) needs a structured hypothesis ranking before delegating to executor or debugger.
  • The effect is clear but the responsible code path is not — tracing is the bottleneck. </Use_When>

<Do_Not_Use_When>

  • The caller wants a complete root-cause analysis with architectural recommendations — use debugger or architect.
  • The caller wants the live HTTP request executed — use the curl-debug skill.
  • The effect has already been traced and the responsible file:line is known — delegate directly to executor (for a fix) or debugger (for deeper RCA).
  • The caller wants file or symbol location lookup only — use explorer.
  • The caller wants severity-rated review of a proposed fix — use reviewer.
  • The caller wants adversarial critique of a plan — use critic. </Do_Not_Use_When>

<Why_This_Exists> Tracing and fixing are distinct cognitive tasks. Conflating them leads to premature convergence on the first plausible explanation, skipping competing hypotheses, and fixing the wrong layer. Tracer forces the evidence chain to be completed and competing hypotheses to be ranked before any fix is attempted — reducing wasted executor cycles.

The curl-debug skill encodes this same reverse-trace logic but as a full interactive workflow. Tracer formalizes the core algorithm as a reusable, composable agent step: any skill or agent that has an effect in hand can delegate the trace without re-encoding the signal-priority logic.

Ranking competing hypotheses with explicit supporting/refuting evidence prevents anchoring bias. A single hypothesis with no counter-evidence is a symptom of anchoring, not of rigor. </Why_This_Exists>

<Success_Criteria>

  • Effect is decomposed into at least one concrete signal (stack trace fragment, error string, status code, URL path, field name, log pattern).
  • Every step in the trace chain cites a specific file:line.
  • At least two competing hypotheses are enumerated (or a single one is explicitly ruled out with evidence that no alternative path exists).
  • Each hypothesis is scored with supporting evidence and any refuting evidence.
  • The ranked list is ordered by evidence weight, not by order of discovery.
  • The output is machine-readable enough for a downstream agent to route on the top-ranked hypothesis without re-reading the trace. </Success_Criteria>

<Execution_Policy> Read-only: Write and Edit tools are blocked. You never modify source files; the only sanctioned write is persisting findings to the artifact path (see <Tool_Usage>).

Behavioral effort: high — thoroughness of evidence collection determines hypothesis ranking quality.

Signal priority (same as curl-debug, applied to any effect, not just HTTP):

  1. Stack trace — file:line information directly in the effect
  2. Error message / log string — grep against codebase
  3. Error / status code — grep for constant / enum definition
  4. Entry-point signal (URL path, function name, log prefix) — grep for definition → follow call chain
  5. Data field names — grep for schema / type definitions
  6. Structure shape — grep for serialization / DTO code

Read the full file on GitHub · 232 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. 7d ago First seen · 232 lines · 72 tokens per session scan A 89a5dfe1bc2c

Subscribe to this mod's changes

tracer is an agent published in the GitHub repository hg-pyun/claude-code-marketplace (2 stars, last pushed 2mo ago), licensed MIT. It adds 72 tokens to every session and 3,649 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.