You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: CHANGELOG.md
+10Lines changed: 10 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -7,6 +7,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
7
7
8
8
## [Unreleased]
9
9
10
+
## [0.10.1] — 2026-05-17
11
+
12
+
### Added
13
+
14
+
-**`regex.code_injection` detector blocks Python subprocess injection attacks.** Catches `__import__('subprocess')` dynamic import bypass, `subprocess.run/Popen` combined with network tools (`curl`, `wget`, `nc`), and `os.system()` with network tools. Covers both the user-instruction vector ("execute this in your code tool") and code tool parameters directly.
15
+
-**`regex.exfil_chain` detector blocks chained tool-abuse with external data exfiltration.** Two patterns: `exfil-chained-to-url` catches "then/and send/upload/forward … http(s)://..." sequences; `exfil-suspicious-path` catches URLs whose path ends in `/collect`, `/exfil`, `/steal`, `/harvest`, or similar collection segments.
16
+
-**`regex.sensitive_file_probe` now blocks privileged file write attempts.** A new `write-etc-privileged` pattern catches write-intent verbs (`write`, `append`, `overwrite`, `modify`, etc.) targeted at `/etc/crontab`, `/etc/sudoers`, `/etc/hosts`, or `/etc/cron.d` — the persistence and privilege-escalation paths an agent should never be asked to write.
17
+
-**PII context honeypot via `armor canary pii-context`.** Four new PII canary types (`pii-name-000`, `pii-email-000`, `pii-dob-000`, `pii-sin-000`) added to the catalogue. The name canary is generated at install time as a randomized "Firstname AwesomeMiddle EpicLast" triple (e.g., "Kevin Lightning Dragon") — distinctive enough to stand out in any output, unique per installation. `write_pii_context()` produces a context snippet to inject into the agent's system prompt; when an attacker asks the agent to compile a PII report, the canary scanner catches the known values at output stage. Run `armor canary generate --out ~/.armor/canaries.json && armor canary pii-context --values ~/.armor/canaries.json --out pii-context.txt`, then inject `pii-context.txt` into your system prompt.
18
+
-**13 new eval corpus rows** covering write-to-crontab, Python code injection, and exfiltration chain patterns, plus matching true negatives.
0 commit comments