evidence-hygiene – Data & Science skill for AI agents
Evidence-capture and PoC-redaction discipline for bug-bounty submissions: cookie redaction protocol (which fields to mask, Preview annotation / Burp panel hiding / DevTools workflow), PII black-bar discipline (what to mask in other-user data — names, emails, phones, faces — vs what is safe to leave — usernames, trace I
Install the evidence-hygiene skill with one command
Pick the tool you use, copy the command and paste it in your terminal. The skill lands in that tool's skills folder and is ready in your next session.
Claude Code
mkdir -p ~/.claude/skills && curl -fsSL https://skill.dn1nguoi.com/z/evidence-hygiene-4574b.tgz | tar xz -C ~/.claude/skills
Call it in Claude Code: /evidence-hygiene – then describe the task you want done.
Codex
mkdir -p ~/.codex/skills && curl -fsSL https://skill.dn1nguoi.com/z/evidence-hygiene-4574b.tgz | tar xz -C ~/.codex/skills
Call it in Codex: $evidence-hygiene – then describe the task you want done.
Hermes Agent
mkdir -p ~/.hermes/skills && curl -fsSL https://skill.dn1nguoi.com/z/evidence-hygiene-4574b.tgz | tar xz -C ~/.hermes/skills
Call it in Hermes Agent: /evidence-hygiene – then describe the task you want done.
DeepSeek
mkdir -p ~/.deepcode/skills && curl -fsSL https://skill.dn1nguoi.com/z/evidence-hygiene-4574b.tgz | tar xz -C ~/.deepcode/skills
Call it in DeepSeek: /evidence-hygiene – then describe the task you want done.
SKILL.md content
EVIDENCE HYGIENE — PoC Capture & Redaction Discipline
> Use this skill BEFORE capturing any screenshot, exporting any HAR, or attaching any evidence to a bug-bounty submission. It catches the most common evidence-hygiene mistakes that cause cookies to leak, PII to be shared without consent, or screenshots to be unsuitable for triage.
The core principle: Bug-bounty evidence is meant to convince a triager. Anything beyond that — live cookies, real-user PII, internal trace IDs that aren't useful — should not be in the evidence.
---
1. Two Categories of Sensitive Data
Every PoC artifact (screenshot, HAR, raw HTTP request, terminal transcript) potentially contains data that needs different treatment.
| Category | Examples | Treatment |
|---|---|---|
| Your-account secrets | Session cookies, OAuth tokens, refresh tokens, API keys | Always redact. Even in private bug-bounty platform attachments. Your account, your session — protect it. |
| Other users' PII | Real names, emails, phone numbers, addresses, profile photos, account IDs | Redact unless explicitly demonstrating cross-account impact. Even then, mask faces and minimize the data you display. |
| Triager-useful metadata | Trace IDs (x-datadog-trace-id), request IDs, server timestamps, your test account UID/email, GraphQL operation names, response shapes | Leave visible — these help the triager correlate to logs and reproduce. |
| Test-account passwords (limited use) | Throwaway passwords on a test account (e.g., Testing@5678) | Acceptable in screenshots if you rotate immediately after submission so the value shown is dead. Don't leave real-use passwords in evidence. |
---
2. Cookie Redaction Protocol
2.1 What must be masked
The session cookie value is the highest-value secret in any PoC. Mask:
- The session cookie (
authn,session,sid,__Secure-id, etc. — name varies per target) csrf-tokenif it's bound to your sessionAuthorizationheaders (Bearer tokens, JWT)Cookierequest header values for any session-bearing cookieSet-Cookieresponse header values for any session-bearing cookie
2.2 What's safe to leave visible
- Cloudflare cookies (
__cf_bm,_cfuvid) — these are bot-management, not session-bearing - Analytics cookies (
ajs_anonymous_id,_ga) - Trace correlation IDs (
x-datadog-trace-id,x-request-id) - Server / framework headers (
Server: cloudflare,X-Frame-Options) - Your test account email/UID (per Bugcrowdninja alias section in
bugcrowd-reporting)
2.3 Redaction methods (ranked by practicality)
Method A — Don't capture the cookies in the first place (preferred when possible)
- For DevTools Console PoCs: use
credentials: 'include'so the browser sends cookies automatically. Console output won't echo the cookie. Screenshot the Console output, never the Network tab Headers panel. - For Burp Repeater PoCs: drag the bottom request/response panel divider DOWN to hide the request body before screenshotting. Capture only the Results table for Intruder runs.
Method B — Black-bar in image editor (when capture inevitably includes cookies)
- macOS: Open screenshot in Preview → Tools → Annotate → Rectangle → set fill color to black → drag rectangle over the cookie value → save
- Windows: Use Snip & Sketch's annotationtools or any image editor (Paint.NET, etc.)
- Burp itself: in Burp's Proxy → Match and Replace, you can pre-emptively redact cookie values to placeholder strings before screenshotting
Method C — Find/replace in raw text (for HAR files, terminal transcripts)
- See §4 for the jq commands
2.4 Pre-screenshot checklist
Before clicking Capture:
[ ] Network tab Headers panel is collapsed or out of frame
[ ] Burp's Request panel is hidden behind the divider drag
[ ] No "Copy as cURL" output is visible on screen
[ ] DevTools Application → Storage → Cookies tab is closed
[ ] Browser URL bar doesn't show a session token in query string (rare but possible)After capturing:
[ ] Open the screenshot at full resolution before saving
[ ] Search for the session cookie name substring in any visible text — if present, redact
[ ] Search for the literal first 6 chars of your cookie value — if present, redact
[ ] Compare to the previous PoC screenshot in the same engagement — same redaction discipline---
3. PII Black-Bar Protocol
When a PoC necessarily exposes another user's data (e.g., demonstrating IDOR by showing the victim's email in an attacker-session response), redact the actual PII even in private attachments.
3.1 What to mask (other-user data)
- First name, last name (full or partial)
- Email address (mask the local part; can leave domain if non-identifying)
- Phone number (mask the last 7 digits, optionally leave country code)
- Physical address (mask everything below city)
- Date of birth (mask the year, optionally the month)
- Government IDs (SSN, passport — mask everything)
- Profile photos / face images (black-bar the face entirely)
- Account IDs that the user could correlate to public profiles
3.2 What to leave visible (proves the bug, not the user)
- The fact that the field was returned (the JSON key name)
- The shape / type of the field (
"first_name": "<REDACTED>") - Your own (attacker session's) UID / email — this proves cross-account
- The endpoint URL and request method
- The trace ID
3.3 Worked example — IDOR PoC body
Bad (leaks victim's full PII):
{"data":{"contact":{"first_name":"Nadene","last_name":"Afton","email":"[email protected]","phone":"+1-555-867-5309"}}}Good (proves the bug, masks the PII):
{"data":{"contact":{"first_name":"<REDACTED — real first name>","last_name":"<REDACTED — real last name>","email":"<REDACTED>@example.com","phone":"<REDACTED>"}}}In screenshot form, black-bar each value with a rectangle annotation labeled "REAL PII REDACTED" if there's space.
3.4 In the report body
Reference the redaction explicitly:
## Proof of Concept
The screenshot below demonstrates the IDOR. The attacker session (uid 12345678) successfully retrieves the victim's profile data (uid 99887766). **Real PII fields in the response are masked with black rectangles to limit unauthorized exposure of victim data, per responsible-disclosure hygiene.** The unredacted response is available privately on request.This signals the triager that you're disciplined and gives them a clear path to the unredacted version if they need it for verification.
---
4. HAR File Sanitization
HAR (HTTP Archive) files are JSON dumps of network traffic with full request/response bodies and headers. They include cookies, auth tokens, and any PII that was in transit.
4.1 Generate the HAR
Chrome DevTools → Network tab → right-click anywhere in the request list → "Save all as HAR with content"
4.2 Sanitize before attaching
Use jq to strip sensitive headers. Save this as a shell function or one-liner you can re-use:
sanitize_har() {
local input="$1"
local output="${1%.har}.sanitized.har"
jq '
.log.entries |= map(
(.request.headers |= map(
if .name | ascii_downcase | IN("cookie", "authorization", "x-csrf-token") then .value = "<REDACTED>" else . end
)) |
(.response.headers |= map(
if .name | ascii_downcase | IN("set-cookie") then .value = "<REDACTED>" else . end
)) |
(.request.cookies |= map(.value = "<REDACTED>")) |
(.response.cookies |= map(.value = "<REDACTED>"))
)
' "$input" > "$output"
echo "Sanitized: $output"
}Usage:
sanitize_har /path/to/exported.har
# Output: /path/to/exported.sanitized.har4.3 Verify before attaching
# Check that no Cookie or Authorization values are leaking
grep -i 'authn\|"cookie"\|authorization' /path/to/exported.sanitized.har | head -20If you see your real cookie value in the output, the sanitization missed something — fix the jq filter for that specific field name.
4.4 Remove other-user PII (if applicable)
If the HAR captured cross-account data (e.g., during an IDOR demo), additionally strip the response body fields that contain victim PII. Add to the jq filter:
(.response.content.text |= (
if . then
(fromjson? // .) | tostring | gsub("real.first.name.example"; "<REDACTED>")
else . end
))Customize the gsub patterns to your specific captured data.
---
5. Burp Suite Screenshot Hygiene
5.1 Repeater (single request demo)
1. In the Repeater request panel, the Cookie header is on its own line — drag the panel divider DOWN to hide everything below the request line / target line
2. Or: temporarily delete the Cookie header text from the Repeater pane (it doesn't affect the original captured request) before screenshotting, then restore
3. Capture only the response panel (right side) showing the JSON / HTML response that demonstrates the bug
5.2 Intruder Results table (rate-lim
… (full content is in the download package)
Related skills
- competitor-profiling – When the user wants to research, profile, or analyze competitors from their URLs. Also…
- clinical-reports – Creates safety-bounded draft structures and runs local deterministic checks for clinical…
- ai-ml – AI and machine learning workflow covering LLM application development, RAG…
- bioinformatics – Gateway to 400+ genomics and computational biology skills.
- consciousness-council – Structures a multi-perspective council exercise for decisions, research trade-offs, and…
- dask – Scales pandas, NumPy, and custom Python research workflows beyond memory or across…
- ai-red-teaming – Run structured AI red team exercises for jailbreak resistance, data exfiltration risk…
- competitive-landscape – Analyze competition, identify differentiation opportunities, and develop winning market…