Changelog¶

All notable changes to the HYDRA × PDX project.

v3.0 — 2026-04-10¶

65+ built-in commands — expanded from 39 to 65+ (40 core + 25 network/process/system)
9-step command pipeline — sanitize → guard → expand → split → pipes → classify → execute → mutate → log
Side-effect registry — apt install now tracked realistically; which, type, and subsequent tool usage work coherently for 25+ packages
bot_dropper classification — new SessionClassifier category for wget/curl-to-sh dropper patterns
Live classification — classify_live runs during the session, not just post-hoc; enables response detail adjustment for signal sessions
Graceful shutdown — SIGINT/SIGTERM/SIGHUP handling with priority-ordered cleanup callbacks
State module — state.init_state() locks project_root/logs_dir/data_dir/config_path as absolute paths, removing cwd dependency

Pipeline Orchestrator — full automation: one-shot, watch, and auto modes for continuous feedback generation
FeedbackEngine — attacker profiling via behavioral clustering, aggregate metric computation, and prompt patch generation
LureGenerator — translates metrics into concrete VFS changes (new decoy files, updated credentials, enriched .bash_history)
FeedbackStore v1.1.0 — new schema with attacker profiles, aggregate metrics, lure recommendations, and prompt patches
Migration system — v1_0_0_to_v1_1_0.py handles feedback.yaml schema upgrades

Full French translation via mkdocs-static-i18n
New pages: Feedback loop, Troubleshooting, FAQ, Changelog, Contributing
Corrected builtin count — 39 → 65+ throughout documentation
Documented undiscovered features — side-effect registry, 9-step pipeline, bot_dropper classification, graceful shutdown

PromptGuard — silent prompt injection detection (0.0–1.0 scoring)
3 personas — fintech_trading, crypto_validator, corp_ad
Virtual filesystem — Copy-on-Write per session
Anti-fingerprinting — SSH banner, cgroup, auth delay, kernel coherence, Kinsing path normalization
Session classifier — bot_ephemeral, bot_exec_scanner, bot_recon, likely_human

DataRouter — dual-use classification (defensive + offensive)
7 training generators — SFT, DPO, RAFT, ReAct, CoT, Chain, JS
Quality pipeline — deduplication, filtering, curriculum ordering, temporal decay
Delta Vector 16D — 16-dimension scoring for each observation
Burp Suite bridge — Java extension + Python proxy
4-tier multi-model router — 7B → 32B → Anthropic API → fallback
8 data collectors — NVD, ExploitDB, OWASP, MITRE ATT&CK, Nuclei, CWE, RFC, man pages