agentleFS
saivarun3407saivarun3407/DFIR-AI/.claude/skills/accuracy-report/SKILL.md

accuracy-report

Skill0 starsChanged 6 months agoApache-2.0View on GitHub

Generates the honest accuracy report — true positives, false positives, false negatives, hallucinations, acknowledged gaps. Required hackathon deliverable. Honesty wins points; inflated claims lose them.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Hallucination corpus result: Run on the 3 deliberately-clean evidence dumps.

    line 29
  2. Very specific:

    Every gap_acknowledged event from the chain.

    line 32
  3. Very specific:

    Every self_correction event from the chain.

    line 35
  4. Very specific:

    Every tool_failure. Don't hide.

    line 38
  5. Distinctive:

    Document any non-determinism source (LLM stochasticity, tool timing, etc.)

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Accuracy Report

The hackathon judges this under "IR Accuracy". Brutal honesty scores higher than polished optimism — judges are senior DFIR practitioners who detect inflated claims instantly.

Required Sections

1. Run Metadata

  • Case ID, evidence inventory (filenames + sha256)
  • Model version, tool versions
  • Total findings, breakdown by confidence enum

2. Confusion Matrix (vs ground truth)

Ground Truth: ThreatGround Truth: Benign
Agent reported: ThreatTPFP (false positive)
Agent reported: BenignFN (missed)TN

For each: count + sample finding IDs.

3. Hallucination Tally

A hallucination is a claim with no valid pin OR a pin whose excerpt doesn't match evidence. The MCP server prevents most by schema enforcement, but list any caught + how (verifier dissent, manual review, etc.).

Hallucination corpus result: Run on the 3 deliberately-clean evidence dumps. Expect zero findings. Report any claims made on these — each is a hallucination.

4. Acknowledged Gaps

Every gap_acknowledged event from the chain. These count POSITIVELY — judges see honest discipline.

5. Self-Correction Events

Every self_correction event from the chain. Show the agent caught its own mistakes mid-run.

6. Tool Failures

Every tool_failure. Don't hide. Show what didn't work and how the agent handled it (gap_acknowledged vs alternative tool).

7. Time + Cost

  • Wall-clock per case
  • Total tokens (input + output)
  • Estimated $ at current Claude pricing
  • Time-to-first-finding

8. Reproducibility Statement

  • "This run is reproducible via ./scripts/replay.sh <case_id>"
  • Document any non-determinism source (LLM stochasticity, tool timing, etc.)
  • Provide expected drift envelope (e.g., "≤2 findings drift run-to-run on identical evidence")

Template

Output to /output/accuracy-report.md. Use this skeleton:

# Accuracy Report — Case <CASE_ID>

## Run Metadata
- Started: <ts>
- Completed: <ts>
- Wall clock: <duration>
- Evidence: <list with sha256>
- Model: <model_id>
- Tool versions: <yaml>

## Findings Summary
- Confirmed: N
- Inferred: N
- Uncertain: N
- Unknown / gaps acknowledged: N

## Confusion Matrix
| | Threat (truth) | Benign (truth) |
|---|---|---|
| Threat (agent) | <TP> | <FP> |
| Benign (agent) | <FN> | <TN> |

Precision: <%>  | Recall: <%>  | F1: <%>

## Hallucinations
<count> caught. List with finding_id, mechanism, resolution.

## Hallucination Corpus Result
- clean-001.dmp: 0 findings ✓
- clean-002.img: 0 findings ✓
- clean-003.tar: 0 findings ✓

## Acknowledged Gaps
<count> total. Examples:
- "<scope>" — <reason>

## Self-Corrections
<count> total. Examples:
- finding_id <X>: original "<...>" → corrected "<...>" — verifier caught <delta>

## Tool Failures
<count>. Each handled by <gap_acknowledged | alternative tool | retry>.

## Cost
- Tokens in: <N>
- Tokens out: <N>
- Estimated cost: $<X>

## Reproducibility
- Run via: `./scripts/replay.sh <CASE_ID>`
- Drift envelope: ≤<N> findings between runs on identical evidence
- Stochastic surfaces: LLM sampling at temperature <T>, otherwise deterministic

Scoring Logic For Self-Assessment

BehaviorScore Direction
High TP rate+
Low FP rate+
Acknowledged gaps documented+
Self-corrections present+
Honest reporting of FNs and tool failures+
Inflated claims (TP without verifier)−
Hidden failures−
Hallucinations not caught−

The point is to be the team that judges trust, not the team with the prettiest claims.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.