agentleFS
medy-gribkovmedy-gribkov/arcana/skills/bugbounty-report/SKILL.md

bugbounty-report

Skill1 starsChanged 8 months agoApache-2.0View on GitHub

Builds and verifies a bug bounty or vulnerability report before submission (HackerOne and similar) with kill-or-keep checks on the finding, the report body and PoC package in platform format, and a compliance check against the program's public policy. Use when writing, formatting or pre-submit auditing a security finding, its PoC zip or its report text, or when someone says bug bounty report, HackerOne report, vulnerability report, security disclosure, PoC package or pre-submit review. For reviewing your own code for vulnerabilities use security-review instead.

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:

    Run this before any finding goes to a bounty platform.

    line 9
  2. Very specific:

    Run the invalidation gauntlet: prove the finding is real, reachable and submittable, or kill it (references/verification.md).

    line 10
  3. Very specific:

    Write the report and PoC to platform format (references/golden-format.md).

    line 11
  4. Distinctive:

    Never submit for the researcher.

    line 15
  5. Very specific:

    Write (references/golden-format.md): the platform's section order, fenced code with language hints, verbatim captures, one version table, …

    line 33

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Bug bounty report: build and verify to golden standard

Run this before any finding goes to a bounty platform. Four jobs, in order:

  1. Run the invalidation gauntlet: prove the finding is real, reachable and submittable, or kill it (references/verification.md). A clean kill, or a downgrade to conditional or informative, is a valid result.
  2. Write the report and PoC to platform format (references/golden-format.md).
  3. Clear the compliance gates from the platform's and the program's own public policies (references/compliance.md).
  4. Hand off field by field for the researcher to paste, verify and submit.

Never submit for the researcher. Never log into, or scrape, the researcher's bounty account or any auth-gated page. The researcher files from their own account, checks the review screen and clicks submit. Walk the hand-off one field at a time and wait for confirmation.

HackerOne basics (static, no need to re-research)

The researcher submits a report to a program. The program's triager validates it, sets the final severity and pays by severity. A submission holds: Title (vulnerability plus asset), Weakness (CWE), Severity (CVSS vector and score), Asset (in scope), Description, Impact (its own field, concrete and bounded), Attachments (the PoC zip, often required for a bounty). Disclosure is program-controlled: hold the report confidential until the program approves. Submit through one channel only (the platform, not also a security email or GHSA).

Operating rules

  • Nothing is trusted until verified. Every source, version and behavior claim maps to a file:line, a command output or a primary-source URL.
  • Platform structure is static. Program facts (scope tier, CVSS version, reward floor, signal caps, disclosure setting) and platform policies are volatile: re-verify them from current public sources on every submission and date them. A fetch that returns only a JavaScript shell is a method failure, not proof the data is private; try another public source.
  • Sourcing order for a needed fact: public web, the codebase, local static notes (which can be stale), the researcher as last resort.
  • Match severity to evidence: the accurate CVSS vector, no inflation, no sandbagging. The program sets the final severity.
  • The PoC must reproduce from a clean state using only its own files. Test it; do not assume it.
  • Voice is terse and factual (references/golden-format.md). No hedges, no severity adjectives, no dashes, no emojis.

Workflow

  1. Gauntlet (references/verification.md): run gates G0 to G9 in order. Each returns KEEP or KILL; one KILL means do not submit, and record the kill honestly in the evidence ledger.
  2. Write (references/golden-format.md): the platform's section order, fenced code with language hints, verbatim captures, one version table, primary-source references.
  3. Scan the package: extract the PoC zip into an empty folder and run node "${CLAUDE_SKILL_DIR}/scripts/scan-package.mjs" <folder>. It flags personal emails, secrets (values are never printed), local user paths, non-reserved endpoints, dashes and emojis, and exits 1 on any finding. Review each hit: a real production endpoint in a script is a fix; a documented URL in a README may be fine. Scan the package, not the report text, since the report legitimately cites URLs.
  4. Comply (references/compliance.md): run the checklist, re-verify program facts, run a duplicate check against public advisories.
  5. Hand off: emit the nine submission fields in order (Asset, Weakness, Severity, Title, Affected versions, Affected file permalink, Description, Impact, Attachment), each as its own copy-paste block. Wrap Description and Impact in four-backtick fences so their inner code blocks survive the paste. State the human-only gates: the researcher confirms the CVSS version and scope on the current public program page, checks the rendered review screen, attaches the zip and submits.

Deliverables

  • The report split into the platform's fields, each copy-paste ready.
  • A PoC package that reproduces from clean and contains no personal data, secrets or local paths.
  • A short compliance verdict: passed items, and the human-only gates the researcher must clear before submitting.

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.

Reviews can't be read right now.

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.