Get card
Download
targeted-debugger@1.1.0.yamlClone
npx -y darkprint clone targeted-debugger@1.1.0There is no repository and no history behind a card. Download hands you the document as it stands, and Clone fetches the same document by name.
Turn one run's failure evidence into the narrowest patch that addresses it, and stop once the attempt cap is spent or the evidence stops changing.
used in 1 blueprint
Specification
127 words · handed to the agentYou are given the evidence from a run that failed: the assertions that did not hold, the stack traces underneath them, and the expected value beside the one actually obtained. Find the smallest change to the existing source that accounts for those failures and emit it on patch a patch, not a rewrite, leaving every part of the code the evidence does not implicate exactly as it is. Do not special-case the literal values in the trace: a change that satisfies only the example in front of you has fixed nothing, and it will come straight back. Stop after three attempts, and stop earlier if two consecutive runs hand you the same evidence, that means you are circling, and another pass costs a round without buying information.
Interfaces
1 in · 1 outInputs
1| Name | Data type | Required | Description |
|---|---|---|---|
| evidence | report | required | One run's failure output, assertions, traces, expected beside obtained. |
Outputs
1| Name | Data type | Description |
|---|---|---|
| patch | code | The narrowest change that accounts for the evidence, ready to be re-run. |
Dependencies
1The upstream nodes this card expects to receive from. Whenever a blueprint pins this card, each name is checked against a real edge in that graph. A name without a link is not a published card; it refers to a node inside some graph.
Card values
19 declaredWho the node is. The id is the key the DOT pins.
- id
- targeted-debugger
- name
- Targeted Debugger
- type
- agent
- phases
- Debugging
What it does, and the prose the agent is handed when the graph runs.
- action25 words
- Turn one run's failure evidence into the narrowest patch that addresses it, and stop once the attempt cap is spent or the evidence stops changing.
- spec127 words
- You are given the evidence from a run that failed: the assertions that did not hold, the stack traces underneath them, and the expected value beside the one actually obtained. Find the smallest change to the existing source that accounts for those failures and emit it on
patcha patch, not a rewrite, leaving every part of the code the evidence does not implicate exactly as it is. Do not special-case the literal values in the trace: a change that satisfies only the example in front of you has fixed nothing, and it will come straight back. Stop after three attempts, and stop earlier if two consecutive runs hand you the same evidence, that means you are circling, and another pass costs a round without buying information.in full above - model
- claude-opus-5
- agent
- Debugger
- skill
- skills/targeted-debugger.md
- tools
- none
- mcp
- filesystem
- params
- max_retries: 2
What arrives, what leaves, which nodes it expects to hear from, and what may not.
- inputs
- evidence : report
- outputs
- patch : code
- dependencies
- acceptance-tester
- cannot
- acceptance-criteria
- will_not
- special-case the literal values in a trace
The keys the static analysis reads. Nothing here instructs the agent.
- risk_markers
- none
- notes180 words
max_retries: 2is the declared cap for thetester -> debugger -> testercycle: two attempts beyond the first, three patches in all, which is the "stop after three attempts" the spec above hands the agent. Engine spec §2.6 counts the attempts that follow the initial execution rather than the executions themselves, so the cap that reads as three is written as two, and the key is spelled the way Attractor spells it. It is load-bearing twice over. Drop it and doc 3 §4.1 charges the blueprint 1.5 for an unbounded loop. It also bounds a slower leak doc 2 §5.5 names: every iteration reveals another slice of the acceptance surface through the failure messages, so the cap limits how much of the criteria this node can reconstruct by accumulation.acceptance-criteriaincannotcloses the fast version of the same leak. Doc 2 §5.5 hands this node the evidence of a failure and withholds the criteria themselves, and because that entry names a data type in the ontology the resolver enforces it: an edge carrying the criteria in raisesbundle/prohibition-violated.
The card's own version, and who wrote it.
- version
- 1.1.0
- author
- autogen
- provenance
- not stated
Version history
2 versions published- targeted-debugger@1.1.0currentsha256:1a0982309df4431d11e32a7856979dc3d6fcfa3a5bdff4acf8bf9bfb132b3bfc
pinned byStarter Software Factory
autogen/starter-software-factoryminor1.0.0 → 1.1.0something was added; old pins still resolve- parameter
max_retrieswas added - parameter
max_iterationswas removed noteschanged
- parameter
- targeted-debugger@1.0.0sha256:60dce0fc202adc9c5e3c144754c34d4ee230aa1d28df9d30d64a84a70d505c73
No blueprint pins this exact version.
A digest is a fingerprint (SHA-256) of the card's content, computed without the author and provenance fields. The same card from two people gets the same digest; any edit gets a new one.
Community notes (0)
No notes yet.
Nobody has posted about this node card yet.
Sign in to post a note.