[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EAGQRYS6RFDCC0XYXVC0T3":3,"skill-same:01M7EAGQRYS6RFDCC0XYXVC0T3":110},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":52,"commitSha":53,"skillDirectory":54,"forksCount":55,"repositoryHtmlUrl":56,"repositoryHomepageUrl":6,"ownerHtmlUrl":57,"ownerBlogUrl":6,"ownerDescription":58,"updatedAt":59,"install":60,"resources":73},"01M7EAGQRYS6RFDCC0XYXVC0T3","web-pentest",null,"Authorized web pentest: recon, proof-based exploits, report.","sanjeevafk\u002Fagent-skills","sanjeevafk","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F230678117?v=4","01M7C540EFY46ETG1NNWKS5X3Q","Sanjeev","01M7C540F366J5C1D587BMKH9Q",0,"MIT","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: web-pentest\ndescription: \"Authorized web pentest: recon, proof-based exploits, report.\"\nversion: 1.0.0\nauthor: Teknium (teknium1), Hermes Agent\nlicense: MIT\nplatforms: [linux, macos]\ncategory: security\ntriggers:\n  - \"pentest [URL]\"\n  - \"pentest this app\"\n  - \"penetration test [URL]\"\n  - \"security test this web app\"\n  - \"test [URL] for vulnerabilities\"\n  - \"find vulns in [URL]\"\n  - \"OWASP test [URL]\"\ntoolsets:\n  - terminal\n  - web\n  - browser\n  - file\n  - delegation\nmetadata:\n  hermes:\n    tags: [Security, Pentest, Web, Recon]\n    related_skills: []\n---\n\n# Web Application Penetration Testing\n\nA phased pentesting workflow for running web applications. Adapted from\nShannon's pipeline (Keygraph, AGPL — concepts only, no code borrowed).\nBuilt around three rules:\n\n1. No exploit, no report — every finding requires reproducible evidence.\n2. Bounded scope — every active request goes against a target the operator\n   pre-declared. Off-scope hosts are refused.\n3. Bypass exhaustion before false-positive dismissal — a \"blocked\" payload\n   is not a clean bill of health until you've tried the bypass set.\n\n---\n\n## ⚠️ Hard Guardrails — Read Before Every Engagement\n\nViolating any of these invalidates the engagement and may be illegal.\n\n1. **Authorization gate.** Before the first active scan in a session, you\n   MUST confirm with the user, in writing, that they own or have written\n   authorization to test the target. Record the acknowledgement in\n   `engagement\u002Fauthorization.md` (see template). No acknowledgement → no\n   active scanning. Reading public pages with `curl` is fine; sending\n   payloads is not.\n\n2. **Scope allowlist.** Maintain `engagement\u002Fscope.txt` — one hostname or\n   CIDR per line. Every `nmap`, `curl`, `whatweb`, browser navigation, or\n   payload-bearing request MUST be against an entry in scope. If a target\n   redirects you off-scope (3xx to a different host, a link in HTML),\n   STOP and confirm with the user before following.\n\n3. **No production systems without paper.** If the user hasn't told you\n   \"yes, prod is in scope and I have written sign-off,\" assume not. Default\n   targets are staging, local docker, dedicated test instances.\n\n4. **Cloud metadata is off by default.** Do not probe `169.254.169.254`,\n   `metadata.google.internal`, `100.100.100.200`, `[fd00:ec2::254]`, or\n   equivalent unless the engagement explicitly includes SSRF-to-metadata\n   as a goal AND the target is one you control. The agent's browser tool\n   can reach these from inside your own infrastructure — don't.\n\n5. **Destructive payloads need approval.** SQLi payloads that DROP\u002FDELETE,\n   filesystem-write SSTI, command injection with `rm`\u002F`shutdown`\u002F`mkfs`,\n   anything that mutates beyond a single test row → ASK FIRST. The\n   `approval.py` system catches some; don't rely on it alone.\n\n6. **Aux-client leakage risk (Hermes-specific).** This skill produces\n   sessions full of SQLi\u002FXSS\u002FRCE payloads, captured credentials, JWT\n   tokens. Hermes' compression and title-generation paths replay history\n   through the auxiliary client (often the main model). Anything sensitive\n   you write to the conversation can leave the box on the next compress.\n   Mitigation:\n   - Redact captured tokens\u002Fcredentials to the LAST 6 CHARS before logging\n     them in any message. Full values go to `engagement\u002Fevidence\u002F` files,\n     never into chat history.\n   - If the engagement is sensitive, set `auxiliary.title_generation.enabled: false`\n     in `~\u002F.hermes\u002Fconfig.yaml` for the session.\n\n7. **Rate limit yourself.** Default 200ms between active requests against\n   any single host. The recon-scan.sh script enforces this. Don't bypass\n   it without operator approval.\n\n8. **Authority of the report.** This skill produces a security\n   assessment, not a \"PASS.\" Even a clean run is \"no exploitable issues\n   FOUND in scope X within time T using methods Y\" — not \"the application\n   is secure.\" Mirror that language in the report.\n\n---\n\n## Phase 0: Engagement Setup\n\nBefore any scanning happens, create the engagement directory and\nauthorization acknowledgement.\n\n```bash\nENGAGEMENT=engagement-$(date +%Y%m%d-%H%M%S)\nmkdir -p \"$ENGAGEMENT\"\u002F{evidence,findings,reports}\ncd \"$ENGAGEMENT\"\n```\n\n1. **Ask the user (verbatim):**\n   > \"Confirm: (a) the target URL is [X], (b) you own this application\n   > or have written authorization to test it, and (c) the engagement\n   > may run for up to [N] hours starting now. Reply 'authorized' to\n   > proceed.\"\n\n2. **Wait for explicit `authorized` response.** Any other answer means STOP.\n\n3. **Record authorization** to `engagement\u002Fauthorization.md` using the\n   template in `templates\u002Fauthorization.md`. Include:\n   - Target URL(s) and IP(s)\n   - Authorization basis (ownership \u002F written authz from $name)\n   - Engagement window\n   - Out-of-scope items (production, third-party services, etc.)\n   - Operator name (the user driving this session)\n\n4. **Build scope.txt:**\n   ```\n   localhost\n   127.0.0.1\n   staging.example.com\n   192.168.1.0\u002F24    # internal lab only, with operator OK\n   ```\n\n5. **Read** `references\u002Fscope-enforcement.md` before issuing the first\n   active request — that doc has the host-extraction rules you apply\n   to every command\u002FURL before it goes out.\n\n---\n\n## Phase 1: Pre-Recon (Code Analysis, optional)\n\nSkip if no source access (black-box engagement).\n\nIf you have read access to the application source:\n\n1. **Map the architecture** — framework, routing, middleware stack\n2. **Inventory sinks** — every `execute(`, `os.system(`, `eval(`,\n   template render, file read\u002Fwrite, redirect target\n3. **Map auth** — session cookie vs JWT, OAuth flows, password reset,\n   privileged endpoints\n4. **Identify trust boundaries** — what's authenticated, what's not,\n   what comes from `request.*`\n5. **Backward taint** from each sink to a request source. Early-terminate\n   when proper sanitization is found (parameterized queries, allowlists,\n   `shlex.quote`, well-known escapers).\n\nOutput: `evidence\u002Fpre-recon.md` — architecture map, sink inventory,\nsuspected vulnerable code paths.\n\nThis is OFFLINE work. No traffic to the target.\n\n---\n\n## Phase 2: Recon (Live, Read-Only)\n\nMaps the attack surface. All requests are GETs of public pages, no\npayloads yet. Still scope-bounded.\n\n1. **Verify scope.** Resolve every target hostname → IP. Confirm IPs are\n   in scope (avoids the \"DNS points somewhere unexpected\" trap).\n\n2. **Network surface** (only if scope permits port scanning):\n   ```bash\n   nmap -sT -T3 --top-ports 100 -oN evidence\u002Fnmap.txt $TARGET\n   ```\n   Use `-T3` (default), not `-T4\u002F-T5`. Stealthier and avoids tripping\n   IDS\u002FIPS in shared environments.\n\n3. **Tech fingerprint:**\n   ```bash\n   whatweb -v $TARGET_URL > evidence\u002Fwhatweb.txt\n   curl -sIk $TARGET_URL > evidence\u002Fheaders.txt\n   ```\n\n4. **Endpoint discovery:**\n   - Crawl the app with `browser_exec` (`new_tab`\u002F`goto_url`, collect\n     links and images via `js()`, follow links).\n   - Inspect `robots.txt`, `sitemap.xml`, `.well-known\u002F*`.\n   - Capture XHR\u002Ffetch calls with `browser_exec`:\n     `js(\"performance.getEntriesByType('resource').map(e => [e.initiatorType, e.name])\")`.\n\n5. **Auth surface:** Identify login, registration, password reset,\n   session cookie names, token formats. Do NOT send credentials yet —\n   just observe.\n\n6. **Correlate with pre-recon** (if you have source). For each\n   `evidence\u002Fpre-recon.md` finding, mark whether the live surface\n   confirms it's reachable.\n\nOutput: `evidence\u002Frecon.md` — endpoints, technologies, auth model,\ninput vectors.\n\n---\n\n## Phase 3: Vulnerability Analysis\n\nOne delegate_task per vulnerability class. Each agent reads\n`evidence\u002Frecon.md` (+ `evidence\u002Fpre-recon.md` if present), produces\n`findings\u002F\u003Cclass>-queue.json` using `templates\u002Fexploitation-queue.json`.\n\nUse `delegate_task` with these focused subagents (parallel where possible):\n\n| Class | Goal | Reference |\n|-------|------|-----------|\n| `injection` | SQLi, command, path traversal, SSTI, LFI\u002FRFI, deserialization | `references\u002Fvuln-taxonomy.md` (slot types) |\n| `xss` | Reflected, stored, DOM-based | `references\u002Fvuln-taxonomy.md` (render contexts) |\n| `auth` | Login bypass, JWT confusion, session fixation, OAuth flaws | `references\u002Fexploitation-techniques.md` |\n| `authz` | IDOR, vertical\u002Fhorizontal escalation, business logic | `references\u002Fexploitation-techniques.md` |\n| `ssrf` | Internal reachability, metadata, protocol smuggling | Skip metadata unless explicitly authorized |\n| `infra` | Misconfig, info disclosure, default creds, exposed admin | `references\u002Fexploitation-techniques.md` |\n\nEach queue entry has: id, vuln class, source (file:line if known),\nendpoint, parameter, slot type, suspected defense, verdict\n(`identified` \u002F `partial` \u002F `confirmed` \u002F `critical`), witness payload,\nconfidence (0-1), notes.\n\nThe analysis phase doesn't send malicious payloads yet — it stages them.\nThe exploitation phase actually fires them.\n\n---\n\n## Phase 4: Exploitation (Proof-Based, Conditional)\n\nOnly run a sub-agent per class where the analysis queue has actionable\nentries (`identified` or `partial`).\n\nFor each candidate:\n\n1. **Pre-send check** — host in scope? auth gate satisfied? payload\n   approved if destructive?\n2. **Send the witness payload** — minimal proof. SQLi: `' AND 1=1--`\n   then `' AND 1=2--`. XSS: a benign marker like\n   `\u003Csvg\u002Fonload=console.log(\"HERMES-PENTEST-XSS\")>`. Never `alert(1)` in\n   stored XSS — it'll fire for other users in shared environments.\n3. **Verify the witness fires** — for blind injection, use a sleep\n   probe (`SLEEP(5)`) and time the response. For SSRF, use a\n   tester-controlled callback host you own (NOT a public service like\n   webhook.site for sensitive engagements — exfil paths).\n4. **Promote level:**\n   - **L1 Identified** — pattern matched, no behavior change\n   - **L2 Partial** — sink reached, but defense in place\n   - **L3 Confirmed** — payload changed app behavior in observable way\n   - **L4 Critical** — data extracted, code executed, access escalated\n5. **Bypass exhaustion before classifying as FP.** For each candidate\n   that blocks: try at least the bypass set in\n   `references\u002Fbypass-techniques.md` for that class. Only after the set\n   is exhausted may you write `verdict: false_positive`.\n6. **Record evidence** for every L3\u002FL4:\n   - Full request (method, URL, headers, body)\n   - Response (status, headers, relevant body excerpt)\n   - Reproducer command (curl one-liner)\n   - Impact statement\n\nOutput: `findings\u002Fexploitation-evidence.md`\n\n**Redact in evidence files:**\n- Any captured credentials\u002Ftokens → last 6 chars only in chat;\n  full value to `findings\u002Fsecrets-vault.md` (gitignored).\n- Other users' PII → redact.\n- Your test credentials → fine to keep.\n\n---\n\n## Phase 5: Reporting\n\nGenerate the final report using `templates\u002Fpentest-report.md`. Sections:\n\n1. Executive summary\n2. Engagement scope (from `engagement\u002Fscope.txt`)\n3. Authorization (from `engagement\u002Fauthorization.md`)\n4. Findings (L3\u002FL4 only — proof-required). Per finding:\n   - Title, severity (CVSS 3.1), CWE\n   - Affected endpoint(s)\n   - Proof (request + response excerpt)\n   - Reproduction steps\n   - Impact\n   - Remediation\n5. Not-exploited candidates (L1\u002FL2 with notes on what blocked them)\n6. Out-of-scope observations\n7. Methodology \u002F tools used\n8. Limitations and what was NOT tested\n\n**Severity policy:** CVSS only for L3\u002FL4. L1\u002FL2 are \"candidates pending\nverification\" — don't assign CVSS to unverified findings.\n\n---\n\n## When to Stop\n\n- The user revokes authorization.\n- A candidate finding clearly impacts production data and you don't have\n  approval for destructive testing — STOP and ask.\n- The target starts returning 503\u002F429 storms — back off, reconvene with\n  the operator.\n- You discover something *outside* the contracted scope (e.g. an exposed\n  customer database while testing an unrelated endpoint). STOP, document,\n  report to the operator. Do not pivot without explicit approval — that\n  pivot is what makes pentesting illegal.\n\n---\n\n## What This Skill Does NOT Cover\n\n- Network-layer pentesting beyond port scanning (no Metasploit,\n  Cobalt Strike, AD attacks, network protocol fuzzing).\n- Reverse engineering \u002F binary analysis (see issue #383).\n- Source-only static analysis (see issue #382).\n- Active social engineering \u002F phishing.\n- Anything against systems the operator hasn't pre-authorized.\n\nIf the engagement needs any of these, escalate to a professional\npentester. This skill complements professional pentesting; it does\nnot replace it.\n\n---\n\n## Further Reading\n\n- `references\u002Fscope-enforcement.md` — how to bound every active request\n- `references\u002Fvuln-taxonomy.md` — slot types, render contexts, OWASP map\n- `references\u002Fexploitation-techniques.md` — per-class payload patterns\n- `references\u002Fbypass-techniques.md` — common WAF\u002Ffilter bypasses\n- `templates\u002Fauthorization.md` — engagement authorization template\n- `templates\u002Fpentest-report.md` — final report template\n- `templates\u002Fexploitation-queue.json` — per-class finding queue schema\n- `scripts\u002Frecon-scan.sh` — rate-limited nmap+whatweb+headers wrapper\n","# Web Application Penetration Testing\n\nA phased pentesting workflow for running web applications. Adapted from\nShannon's pipeline (Keygraph, AGPL — concepts only, no code borrowed).\nBuilt around three rules:\n\n1. No exploit, no report — every finding requires reproducible evidence.\n2. Bounded scope — every active request goes against a target the operator\n   pre-declared. Off-scope hosts are refused.\n3. Bypass exhaustion before false-positive dismissal — a \"blocked\" payload\n   is not a clean bill of health until you've tried the bypass set.\n\n---\n\n## ⚠️ Hard Guardrails — Read Before Every Engagement\n\nViolating any of these invalidates the engagement and may be illegal.\n\n1. **Authorization gate.** Before the first active scan in a session, you\n   MUST confirm with the user, in writing, that they own or have written\n   authorization to test the target. Record the acknowledgement in\n   `engagement\u002Fauthorization.md` (see template). No acknowledgement → no\n   active scanning. Reading public pages with `curl` is fine; sending\n   payloads is not.\n\n2. **Scope allowlist.** Maintain `engagement\u002Fscope.txt` — one hostname or\n   CIDR per line. Every `nmap`, `curl`, `whatweb`, browser navigation, or\n   payload-bearing request MUST be against an entry in scope. If a target\n   redirects you off-scope (3xx to a different host, a link in HTML),\n   STOP and confirm with the user before following.\n\n3. **No production systems without paper.** If the user hasn't told you\n   \"yes, prod is in scope and I have written sign-off,\" assume not. Default\n   targets are staging, local docker, dedicated test instances.\n\n4. **Cloud metadata is off by default.** Do not probe `169.254.169.254`,\n   `metadata.google.internal`, `100.100.100.200`, `[fd00:ec2::254]`, or\n   equivalent unless the engagement explicitly includes SSRF-to-metadata\n   as a goal AND the target is one you control. The agent's browser tool\n   can reach these from inside your own infrastructure — don't.\n\n5. **Destructive payloads need approval.** SQLi payloads that DROP\u002FDELETE,\n   filesystem-write SSTI, command injection with `rm`\u002F`shutdown`\u002F`mkfs`,\n   anything that mutates beyond a single test row → ASK FIRST. The\n   `approval.py` system catches some; don't rely on it alone.\n\n6. **Aux-client leakage risk (Hermes-specific).** This skill produces\n   sessions full of SQLi\u002FXSS\u002FRCE payloads, captured credentials, JWT\n   tokens. Hermes' compression and title-generation paths replay history\n   through the auxiliary client (often the main model). Anything sensitive\n   you write to the conversation can leave the box on the next compress.\n   Mitigation:\n   - Redact captured tokens\u002Fcredentials to the LAST 6 CHARS before logging\n     them in any message. Full values go to `engagement\u002Fevidence\u002F` files,\n     never into chat history.\n   - If the engagement is sensitive, set `auxiliary.title_generation.enabled: false`\n     in `~\u002F.hermes\u002Fconfig.yaml` for the session.\n\n7. **Rate limit yourself.** Default 200ms between active requests against\n   any single host. The recon-scan.sh script enforces this. Don't bypass\n   it without operator approval.\n\n8. **Authority of the report.** This skill produces a security\n   assessment, not a \"PASS.\" Even a clean run is \"no exploitable issues\n   FOUND in scope X within time T using methods Y\" — not \"the application\n   is secure.\" Mirror that language in the report.\n\n---\n\n## Phase 0: Engagement Setup\n\nBefore any scanning happens, create the engagement directory and\nauthorization acknowledgement.\n\n```bash\nENGAGEMENT=engagement-$(date +%Y%m%d-%H%M%S)\nmkdir -p \"$ENGAGEMENT\"\u002F{evidence,findings,reports}\ncd \"$ENGAGEMENT\"\n```\n\n1. **Ask the user (verbatim):**\n   > \"Confirm: (a) the target URL is [X], (b) you own this application\n   > or have written authorization to test it, and (c) the engagement\n   > may run for up to [N] hours starting now. Reply 'authorized' to\n   > proceed.\"\n\n2. **Wait for explicit `authorized` response.** Any other answer means STOP.\n\n3. **Record authorization** to `engagement\u002Fauthorization.md` using the\n   template in `templates\u002Fauthorization.md`. Include:\n   - Target URL(s) and IP(s)\n   - Authorization basis (ownership \u002F written authz from $name)\n   - Engagement window\n   - Out-of-scope items (production, third-party services, etc.)\n   - Operator name (the user driving this session)\n\n4. **Build scope.txt:**\n   ```\n   localhost\n   127.0.0.1\n   staging.example.com\n   192.168.1.0\u002F24    # internal lab only, with operator OK\n   ```\n\n5. **Read** `references\u002Fscope-enforcement.md` before issuing the first\n   active request — that doc has the host-extraction rules you apply\n   to every command\u002FURL before it goes out.\n\n---\n\n## Phase 1: Pre-Recon (Code Analysis, optional)\n\nSkip if no source access (black-box engagement).\n\nIf you have read access to the application source:\n\n1. **Map the architecture** — framework, routing, middleware stack\n2. **Inventory sinks** — every `execute(`, `os.system(`, `eval(`,\n   template render, file read\u002Fwrite, redirect target\n3. **Map auth** — session cookie vs JWT, OAuth flows, password reset,\n   privileged endpoints\n4. **Identify trust boundaries** — what's authenticated, what's not,\n   what comes from `request.*`\n5. **Backward taint** from each sink to a request source. Early-terminate\n   when proper sanitization is found (parameterized queries, allowlists,\n   `shlex.quote`, well-known escapers).\n\nOutput: `evidence\u002Fpre-recon.md` — architecture map, sink inventory,\nsuspected vulnerable code paths.\n\nThis is OFFLINE work. No traffic to the target.\n\n---\n\n## Phase 2: Recon (Live, Read-Only)\n\nMaps the attack surface. All requests are GETs of public pages, no\npayloads yet. Still scope-bounded.\n\n1. **Verify scope.** Resolve every target hostname → IP. Confirm IPs are\n   in scope (avoids the \"DNS points somewhere unexpected\" trap).\n\n2. **Network surface** (only if scope permits port scanning):\n   ```bash\n   nmap -sT -T3 --top-ports 100 -oN evidence\u002Fnmap.txt $TARGET\n   ```\n   Use `-T3` (default), not `-T4\u002F-T5`. Stealthier and avoids tripping\n   IDS\u002FIPS in shared environments.\n\n3. **Tech fingerprint:**\n   ```bash\n   whatweb -v $TARGET_URL > evidence\u002Fwhatweb.txt\n   curl -sIk $TARGET_URL > evidence\u002Fheaders.txt\n   ```\n\n4. **Endpoint discovery:**\n   - Crawl the app with `browser_exec` (`new_tab`\u002F`goto_url`, collect\n     links and images via `js()`, follow links).\n   - Inspect `robots.txt`, `sitemap.xml`, `.well-known\u002F*`.\n   - Capture XHR\u002Ffetch calls with `browser_exec`:\n     `js(\"performance.getEntriesByType('resource').map(e => [e.initiatorType, e.name])\")`.\n\n5. **Auth surface:** Identify login, registration, password reset,\n   session cookie names, token formats. Do NOT send credentials yet —\n   just observe.\n\n6. **Correlate with pre-recon** (if you have source). For each\n   `evidence\u002Fpre-recon.md` finding, mark whether the live surface\n   confirms it's reachable.\n\nOutput: `evidence\u002Frecon.md` — endpoints, technologies, auth model,\ninput vectors.\n\n---\n\n## Phase 3: Vulnerability Analysis\n\nOne delegate_task per vulnerability class. Each agent reads\n`evidence\u002Frecon.md` (+ `evidence\u002Fpre-recon.md` if present), produces\n`findings\u002F\u003Cclass>-queue.json` using `templates\u002Fexploitation-queue.json`.\n\nUse `delegate_task` with these focused subagents (parallel where possible):\n\n| Class | Goal | Reference |\n|-------|------|-----------|\n| `injection` | SQLi, command, path traversal, SSTI, LFI\u002FRFI, deserialization | `references\u002Fvuln-taxonomy.md` (slot types) |\n| `xss` | Reflected, stored, DOM-based | `references\u002Fvuln-taxonomy.md` (render contexts) |\n| `auth` | Login bypass, JWT confusion, session fixation, OAuth flaws | `references\u002Fexploitation-techniques.md` |\n| `authz` | IDOR, vertical\u002Fhorizontal escalation, business logic | `references\u002Fexploitation-techniques.md` |\n| `ssrf` | Internal reachability, metadata, protocol smuggling | Skip metadata unless explicitly authorized |\n| `infra` | Misconfig, info disclosure, default creds, exposed admin | `references\u002Fexploitation-techniques.md` |\n\nEach queue entry has: id, vuln class, source (file:line if known),\nendpoint, parameter, slot type, suspected defense, verdict\n(`identified` \u002F `partial` \u002F `confirmed` \u002F `critical`), witness payload,\nconfidence (0-1), notes.\n\nThe analysis phase doesn't send malicious payloads yet — it stages them.\nThe exploitation phase actually fires them.\n\n---\n\n## Phase 4: Exploitation (Proof-Based, Conditional)\n\nOnly run a sub-agent per class where the analysis queue has actionable\nentries (`identified` or `partial`).\n\nFor each candidate:\n\n1. **Pre-send check** — host in scope? auth gate satisfied? payload\n   approved if destructive?\n2. **Send the witness payload** — minimal proof. SQLi: `' AND 1=1--`\n   then `' AND 1=2--`. XSS: a benign marker like\n   `\u003Csvg\u002Fonload=console.log(\"HERMES-PENTEST-XSS\")>`. Never `alert(1)` in\n   stored XSS — it'll fire for other users in shared environments.\n3. **Verify the witness fires** — for blind injection, use a sleep\n   probe (`SLEEP(5)`) and time the response. For SSRF, use a\n   tester-controlled callback host you own (NOT a public service like\n   webhook.site for sensitive engagements — exfil paths).\n4. **Promote level:**\n   - **L1 Identified** — pattern matched, no behavior change\n   - **L2 Partial** — sink reached, but defense in place\n   - **L3 Confirmed** — payload changed app behavior in observable way\n   - **L4 Critical** — data extracted, code executed, access escalated\n5. **Bypass exhaustion before classifying as FP.** For each candidate\n   that blocks: try at least the bypass set in\n   `references\u002Fbypass-techniques.md` for that class. Only after the set\n   is exhausted may you write `verdict: false_positive`.\n6. **Record evidence** for every L3\u002FL4:\n   - Full request (method, URL, headers, body)\n   - Response (status, headers, relevant body excerpt)\n   - Reproducer command (curl one-liner)\n   - Impact statement\n\nOutput: `findings\u002Fexploitation-evidence.md`\n\n**Redact in evidence files:**\n- Any captured credentials\u002Ftokens → last 6 chars only in chat;\n  full value to `findings\u002Fsecrets-vault.md` (gitignored).\n- Other users' PII → redact.\n- Your test credentials → fine to keep.\n\n---\n\n## Phase 5: Reporting\n\nGenerate the final report using `templates\u002Fpentest-report.md`. Sections:\n\n1. Executive summary\n2. Engagement scope (from `engagement\u002Fscope.txt`)\n3. Authorization (from `engagement\u002Fauthorization.md`)\n4. Findings (L3\u002FL4 only — proof-required). Per finding:\n   - Title, severity (CVSS 3.1), CWE\n   - Affected endpoint(s)\n   - Proof (request + response excerpt)\n   - Reproduction steps\n   - Impact\n   - Remediation\n5. Not-exploited candidates (L1\u002FL2 with notes on what blocked them)\n6. Out-of-scope observations\n7. Methodology \u002F tools used\n8. Limitations and what was NOT tested\n\n**Severity policy:** CVSS only for L3\u002FL4. L1\u002FL2 are \"candidates pending\nverification\" — don't assign CVSS to unverified findings.\n\n---\n\n## When to Stop\n\n- The user revokes authorization.\n- A candidate finding clearly impacts production data and you don't have\n  approval for destructive testing — STOP and ask.\n- The target starts returning 503\u002F429 storms — back off, reconvene with\n  the operator.\n- You discover something *outside* the contracted scope (e.g. an exposed\n  customer database while testing an unrelated endpoint). STOP, document,\n  report to the operator. Do not pivot without explicit approval — that\n  pivot is what makes pentesting illegal.\n\n---\n\n## What This Skill Does NOT Cover\n\n- Network-layer pentesting beyond port scanning (no Metasploit,\n  Cobalt Strike, AD attacks, network protocol fuzzing).\n- Reverse engineering \u002F binary analysis (see issue #383).\n- Source-only static analysis (see issue #382).\n- Active social engineering \u002F phishing.\n- Anything against systems the operator hasn't pre-authorized.\n\nIf the engagement needs any of these, escalate to a professional\npentester. This skill complements professional pentesting; it does\nnot replace it.\n\n---\n\n## Further Reading\n\n- `references\u002Fscope-enforcement.md` — how to bound every active request\n- `references\u002Fvuln-taxonomy.md` — slot types, render contexts, OWASP map\n- `references\u002Fexploitation-techniques.md` — per-class payload patterns\n- `references\u002Fbypass-techniques.md` — common WAF\u002Ffilter bypasses\n- `templates\u002Fauthorization.md` — engagement authorization template\n- `templates\u002Fpentest-report.md` — final report template\n- `templates\u002Fexploitation-queue.json` — per-class finding queue schema\n- `scripts\u002Frecon-scan.sh` — rate-limited nmap+whatweb+headers wrapper\n",{"name":5,"author":24,"license":15,"version":25,"category":26,"metadata":27,"toolsets":35,"triggers":41,"platforms":49,"description":7},"Teknium (teknium1), Hermes Agent","1.0.0","security",{"hermes":28},{"tags":29,"related_skills":34},[30,31,32,33],"Security","Pentest","Web","Recon",[],[36,37,38,39,40],"terminal","web","browser","file","delegation",[42,43,44,45,46,47,48],"pentest [URL]","pentest this app","penetration test [URL]","security test this web app","test [URL] for vulnerabilities","find vulns in [URL]","OWASP test [URL]",[50,51],"linux","macos","main","4e084e84eb42fc53e144c89c0f7f48b39c8a4b0f","collections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest",1,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills","https:\u002F\u002Fgithub.com\u002Fsanjeevafk","First make it work. Then make it right. Then make it fast.","2026-11-14 13:36:16",{"source":8,"skillName":5,"commands":61,"prompt":71,"zipAvailable":72},[62,65,68],{"tool":63,"command":64},"npx","npx skills add sanjeevafk\u002Fagent-skills --skill web-pentest",{"tool":66,"command":67},"bunx","bunx skills add sanjeevafk\u002Fagent-skills --skill web-pentest",{"tool":69,"command":70},"pnpm","pnpm dlx skills add sanjeevafk\u002Fagent-skills --skill web-pentest","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002FSKILL.md",true,[74,80,84,88,92,98,102,106],{"path":75,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":78,"sourceHtmlUrl":79},"references\u002Fbypass-techniques.md","REFERENCE","参考文档",4645,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Freferences\u002Fbypass-techniques.md",{"path":81,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":82,"sourceHtmlUrl":83},"references\u002Fexploitation-techniques.md",6820,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Freferences\u002Fexploitation-techniques.md",{"path":85,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":86,"sourceHtmlUrl":87},"references\u002Fscope-enforcement.md",4169,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Freferences\u002Fscope-enforcement.md",{"path":89,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":90,"sourceHtmlUrl":91},"references\u002Fvuln-taxonomy.md",4394,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Freferences\u002Fvuln-taxonomy.md",{"path":93,"resourceType":94,"resourceTypeDesc":95,"sizeBytes":96,"sourceHtmlUrl":97},"scripts\u002Frecon-scan.sh","SCRIPT","脚本",3930,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Fscripts\u002Frecon-scan.sh",{"path":99,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":100,"sourceHtmlUrl":101},"templates\u002Fauthorization.md",2408,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Ftemplates\u002Fauthorization.md",{"path":103,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":104,"sourceHtmlUrl":105},"templates\u002Fexploitation-queue.json",1236,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Ftemplates\u002Fexploitation-queue.json",{"path":107,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":108,"sourceHtmlUrl":109},"templates\u002Fpentest-report.md",4666,"https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fcollections\u002Fhermes-agent\u002Fskills\u002Fweb-pentest\u002Ftemplates\u002Fpentest-report.md",[111,118,126,135,143,150],{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"publishedAt":19},"01M7EFHX67M3KKXT1BXVPA5ADT","zarr-python","Chunked N-D arrays for cloud storage (Zarr-Python 3). Compressed arrays, parallel I\u002FO, S3\u002FGCS via fsspec, NumPy\u002FDask\u002FXarray compatible, for large-scale scientific computing pipelines.","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fzarr-python\u002FSKILL.md",[],[],{"skillNo":119,"name":120,"nameZh":6,"description":121,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":122,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":19},"01M7EFFQGM51TZBGRENXY50ZYM","whiteboard-diagrams","Create accurate, simple, hand-drawn whiteboard diagrams from a user's process, system, architecture, learning workflow, comparison, or causal explanation. Use when the user asks for a scribbly whiteboard visual, educational workflow graphic, hand-drawn explainer, or diagram like the bundled reference; ask one focused clarification question when the subject or relationships are missing, then generate and visually verify the final image.","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwhiteboard-diagrams\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwhiteboard-diagrams\u002FSKILL.md",[],[],{"skillNo":127,"name":128,"nameZh":6,"description":129,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":130,"licenseFileUrl":131,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":19},"01M7EFF1CSQNTDMHFK4ZGY6NYK","webapp-testing","Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwebapp-testing\u002FSKILL.md",[],[],{"skillNo":136,"name":137,"nameZh":6,"description":138,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":130,"licenseFileUrl":139,"sourceHtmlUrl":140,"categories":141,"keywords":142,"publishedAt":19},"01M7EFE8106X6C28FT8PA74427","web-artifacts-builder","Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn\u002Fui). Use for complex artifacts requiring state management, routing, or shadcn\u002Fui components - not for simple single-file HTML\u002FJSX artifacts.","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fweb-artifacts-builder\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fweb-artifacts-builder\u002FSKILL.md",[],[],{"skillNo":144,"name":145,"nameZh":6,"description":146,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":147,"categories":148,"keywords":149,"publishedAt":19},"01M7EFCSNZM9G7ZHS6F127Q71X","verification-loop","A comprehensive verification system for Claude Code sessions.","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fverification-loop\u002FSKILL.md",[],[],{"skillNo":151,"name":152,"nameZh":6,"description":153,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":154,"sourceHtmlUrl":155,"categories":156,"keywords":157,"publishedAt":19},"01M7EFCE7EZMKPRYDDNKBQ9EEH","vercel-deploy","Deploy applications and websites to Vercel. Use when the user requests deployment actions like \"deploy my app\", \"deploy and give me the link\", \"push this live\", or \"create a preview deployment\".","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fvercel-deploy\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fsanjeevafk\u002Fagent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fvercel-deploy\u002FSKILL.md",[],[]]