Works with the AI Agents you already use
Free Triager Pro
You wrote it up, attached the proof, picked the severity row, and paid the submission fee.
$14.99
Free Triager Pro
Example session with this skill installed
You can either prompt in your own words, like:
Triage this report [PATH TO REPORT] for the Immunefi program at [PROGRAM URL]. The source is at [PATH OR URL TO SOURCE FILES OF CODEBASE].
Or use a one-liner as below
$free-triager "C:\path\to\report.md" https://immunefi.com/bug-bounty/example/information/ "C:\path\to\codebase"
The skill only requires three things
- The report
- The program page URL
- The Path to the source files of the codebase or GitHub URL (source files are better).
If something is missing, the skill asks you for it.
- Read your context and instructions
- Compiled the free triager pro
I ran the five-phase triage on the synthetic fixture. Verdict: ready_with_changes — submit after changes, Low severity. No files were changed; this was review-only.
PROVENANCE
Program page: Local stand-in at release/demo/fixture/program-page.md for https://immunefi.com/bug-bounty/synthetic-example/information/, opened once at the start. Not a live fetch.
Source: release/demo/fixture/synthetic-target, declared revision label demo-0001. The fixture has no VCS metadata, so demo-0001 is a label, not an immutable commit.
WHY IT HOLDS
• Scope: The page lists exactly one asset, Synthetic Vault src/withdraw.js at demo-0001. The report matches it.
• Root cause: Proven in src/withdraw.js. rescue(now) adds periodOf(now) to the global lockedPeriods Set (lines 27–29); withdraw(user, amount, now) throws whenever that period is present (lines 31–34), before any balance read or mutation. The lock has no account term and is never cleared.
• Impact: One ordinary authorised rescue locks one 24-hour epoch-aligned period (maximum 24 hours, approximately 12 hours on average). The entry persists, but its effect self-releases at the next period boundary. There is no theft or unauthorised state change. This maps directly to the program's Low severity row, "temporary freezing of funds for less than 7 days." No CVSS. A single rescue cannot exceed 7 days; repeated rescues could chain, but that is a different scenario and is not claimed.
• Eligibility: No known issues ("None disclosed"), no duplicate, and not purely informational. The "compromised operator key" exclusion is not triggered because the rescue is ordinary and authorised; no key compromise is needed. The fixture README's "planted defect" note is demo provenance, not a program disclosure.
• PoC requirements: Code plus four clear steps satisfy the program's stated proof-of-concept rule; runnable execution is expressly not required.
BEFORE SUBMITTING
-
Select the program's Low impact row in the dashboard and state it in the Impact section.
-
Make the PoC demonstrate the cross-user claim: deposit for Alice and Bob, call rescue(now), and show both withdrawals thrown in the same period. Use one captured now, an explicit require path, and assert. throws, a Node run command, and pasted output.
-
State who controls now. It is caller-supplied and unvalidated. The freeze holds for callers presenting the true current period, but a caller can bypass it with an unlocked-period timestamp. If users control it, disclose the bypass and narrow the claim; the impact may drop to informational.
-
Add Brief/Intro, Risk Breakdown, Recommendation, and References. Cite line numbers and note that demo-0001 is a label without a commit pin. Tighten the title to "remainder of the current period (max 24h)" and fix the imprecise "not by the rescued account" phrasing. The rescue function has no account parameter; the lock is simply global.
-
For a real submission, resolve authorship and automation truthfully, as the synthetic "No human author" placeholder would trigger that check. Remove the synthetic banner and re-read the live program page.
OPEN ITEMS
• The live page was not fetched; this was a local stand-in.
• There is no immutable commit for demo-0001.
• It remains unclear who supplies now, whether rescue is routine, and where authorisation lives outside the single asset.
• The report's PoC was statically verified, not executed in this invocation. The separate fixture transcript corroborates the throw with the balance unchanged (100 before and after), but this invocation did not execute it.
• Automated-triage readiness is at_risk until items 1–2 and the authorship issue are fixed. Human-review merits are valid at Low. Fee and reward terms are not stated, so fee risk is unknown. Confidence: medium-high.
NEXT ACTION
Make changes 1–4 above, re-read the live page, then submit. The original report was left untouched; a revised copy can be prepared upon explicit request.
Connects securely to your tools. The creator never sees your data.
About this skill
Free Triager Pro
You found something real. You wrote it up, attached the proof, picked the severity row, and paid the submission fee. On Immunefi, automated front-line triage such as Andrew can reject that submission before a human reads its technical merits. The fee is gone either way: known issue, out-of-scope asset, wrong PoC form, a claim that does not hold up against the program rules. Depending on the platform and the tier, what you just paid for a real read-through runs from tens to hundreds of dollars.
Free Triager was built exactly on that loss. It is the same review, run before you pay. It reads the live program rules and the affected source, then checks what a front line checks: is the finding technically supported in the source, in scope under the current rules, does the report and PoC meet the stated requirements, and what specifically could get it rejected or downgraded. If the report is ready, you will know why; if it is not, you get the numbered fixes while they still cost a rewrite instead of a fee.
The four checks
| Check | Question answered |
|---|---|
| Technical validity | Does the behaviour exist in the affected source, and is the attack path realistic and reproducible? |
| Program eligibility | Are the asset, impact, and assumptions covered by the current rules, and is the finding not a disclosed or closed duplicate? |
| Submission readiness | Does the report and its proof of concept meet the platform's stated requirements? |
| Rejection risk | Which specific issues could cause automated or human triage to reject or downgrade the report, and how they are corrected |
The output separates automated-triage readiness from the underlying technical merits. A valid but poorly presented report is not treated as an invalid vulnerability, and a missing source makes a finding unverified rather than false.
Why you should invest in it
- The fee you keep: the check catches known issues, out-of-scope assets, unsupported claims, and wrong PoC form before you pay. Guarded fees run from tens to hundreds of dollars, so one avoided waste earns the price back.
- A verdict you can act on: every conclusion traces to the live program page and to specific lines in the affected source, so you fix what is real or defend what is right with the quotes behind you.
- A fix list instead of another guess: a
ready_with_changesverdict arrives with numbered corrections for impact selection, PoC structure, and missing sections, so the resubmission is the second draft instead of the fifth. - Your report stays yours: a run keeps everything in memory, writes nothing to disk, and writes a revision only when you explicitly ask for one.
- Both platforms handled as they behave: HackenProof's Dual Defence, audit competition, and bug bounty programs carry separate duplicate, PoC, fee, and reward rules; each is treated as its own rule set instead of assuming bug bounty everywhere.
- Smart contracts in any language or virtual machine, web and application targets, blockchain and distributed-ledger implementations, and mixed findings that cross contracts, applications, nodes, bridges, or other system boundaries: the check derives its approach from your report, the program, and the evidence, so no language or framework allowlist ever filters the source.
How a run works
- You supply three inputs: the report, the program URL, and the affected source. The source is a local path or a GitHub repository, tree, or file URL; a GitHub source is read remotely, with no clone. Source access is mandatory.
- A five-phase workflow (prior art, program policy, eligibility gate, technical validation, verdict) verifies the report against the live page and the implementation: repository identity, resolved commit, scoped directories, and correspondence with the program's target. Branch URLs resolve to the immutable commit actually read, and an ambiguous slash-ref boundary is resolved through read-only GitHub metadata rather than guessed.
- You get a verdict (
ready,ready_with_changes,needs_information), severity, automated-triage readiness, human-review merits, and a fix list.
What Pro gives you that the free edition does not
The edition that stays free and MIT-licensed on GitHub remains available. Pro is the edition behind this price: mandatory source verification with revision pinning, read-only slash-ref boundary resolution, HackenProof policy normalisation and hardening, classifier hardening for blocked pages and negated labels, and the single in-memory execution model. Provenance is documented in NOTICE.md inside the package.
What comes next
The adapters are independent by design: a future platform ships as its own adapter beside the Immunefi and HackenProof adapters, without a rewrite of the triage core.
More platforms arrive over time, and the listing price moves with them. The price you buy at is the price you keep; past buyers are never charged when the price changes, so the current price is the cheap side of the listing.
Where it stops
- No live scanning, exploitation, or testing of targets; the tool reviews the report you supply.
- It is an evidence-based assessment, not a guarantee that a platform will accept the report.
- No reachable source means
needs_information; source unavailability never proves a finding invalid. - Formats and claims were designed around Immunefi and HackenProof. Other platforms return
unsupported_platformtoday.
- Reviews reports; does not scan, exploit, or test any target.
- An assessment, not a guarantee that a platform accepts the report.
- Requires a reachable source; otherwise, the run stops with
needs_information. - Supports two platform adapters today: Immunefi and HackenProof. Others return
unsupported_platform. - HackenProof runs require Node 18+; no Node-free path is documented for the helper steps.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 20 days ago
Requires
Permissions
Every run reads the live program page and its linked rules, plus the declared GitHub source through a read-only browser or API access; no content from your machine is uploaded back to the program website. Report and source are read from paths you declare, so file scopes cannot be narrowed without breaking valid triage. Write access exists only for a revised report you explicitly request after seeing the assessment; otherwise, a run keeps all state in memory, and nothing is cached or logged to disk.