Security: what we found, and what we did
Every vulnerability, bug and breach found on this site is published here, ours included. A fixed finding is shown in full; an open one shows only its title, severity and state until it is fixed. The log for every PrompTugal site: promptugal.news/security/. The data: /api/v1/security/findings.json.
Findings
| Id | Finding | Severity | Found | State | Files | Fix | Commit |
|---|---|---|---|---|---|---|---|
| DATA-2026-001 | Links built from fetched URLs kept any scheme (javascript: possible) | medium | 2026-11-18, by AppSec post-review of 1bf3ad4..55b6db3..HEAD (data.b.security@PrompTugal.dinis, Sonnet, checked by data.b@) | fixed | scripts/build/render.mjs | inline() keeps only http(s), site-relative, # and mailto links; anything else becomes # (Event URLs from the agendas and evidence URLs from Perplexity answers reached label links on /events/ and the source pages; a javascript: URL would have run script in the reader's browser on a click. Quotes were already escaped. No such URL was seen in the data.) | 1b8d304 |
| DATA-2026-002 | robots.txt "Disallow: /" was read as allowed by the events collector | medium | 2026-11-18, by AppSec post-review of 1bf3ad4..55b6db3..HEAD (data.b.security@PrompTugal.dinis, Sonnet, checked by data.b@) | fixed | sources/events-pt/collect.mjs | the root disallow now blocks the host (A site banning all crawling would still have been read. None of the hosts read had such a rule.) | 1b8d304 |
| DATA-2026-003 | An event without an http(s) URL could break the site build | medium | 2026-11-18, by AppSec post-review of 1bf3ad4..55b6db3..HEAD (data.b.security@PrompTugal.dinis, Sonnet, checked by data.b@) | fixed | sources/events-pt/collect.mjs, scripts/build/pages.mjs | such events are dropped and counted | 1b8d304 |
| DATA-2026-004 | URL filters missing in discovery, claims, the social collector and the charts | low | 2026-11-18, by AppSec post-review of 1bf3ad4..55b6db3..HEAD (data.b.security@PrompTugal.dinis, Sonnet, checked by data.b@) | fixed | collect/discovery.mjs, collect/claims.mjs, sources/social/collect.mjs, src/assets/viz/viz.js, src/assets/viz/flightmap.js, src/assets/viz/storm-labs.js | non-http(s) URLs dropped or replaced by #; malformed model items skipped | 1b8d304 |
| DATA-2026-005 | The observability credentials folder was created after the credentials file | low | 2026-11-18, by AppSec post-review of 1bf3ad4..55b6db3..HEAD (data.b.security@PrompTugal.dinis, Sonnet, checked by data.b@) | fixed | scripts/obs-data-setup.py | the folder is made 0700 first (the file itself was 0600; one-user container) | 1b8d304 |
| DATA-2026-006 | adsb.lol raw answers, which name private aircraft by hex id and position, went to a public-read vault | medium | 2026-11-18, by a review | open | |||
| DATA-2026-007 | Perplexity answer text is served as evidence and could carry injected instructions for a model that reads it later | low | 2026-11-18, by AppSec post-review of 1bf3ad4..55b6db3..HEAD (data.b.security@PrompTugal.dinis, Sonnet, checked by data.b@) | accepted | scripts/build/discovery.mjs | none: kept as evidence, marked untrusted in the threat model; no injection text found on 18 November | |
| DATA-2026-008 | CI actions pinned by tag, not by commit | low | 2026-11-18, by a review | open |
Threat models
Each automation has a published threat model: what it reads, writes and trusts, what could go wrong, the mitigations. The list: discovery, social signals; the others are being written. Reviewed by security@PrompTugal.dinis.
Report a problem
Found something? Open an issue (for anything that should not be public, say so in the title and leave out the detail; we will ask how to reach you).