Skip to content

Commit c6c81e8

Browse files
Version Packages (#3)
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
1 parent ff4bf12 commit c6c81e8

9 files changed

Lines changed: 24 additions & 31 deletions

.changeset/bash-trim-conditional-dedup.md

Lines changed: 0 additions & 5 deletions
This file was deleted.

.changeset/safeguard-config-redesign.md

Lines changed: 0 additions & 5 deletions
This file was deleted.

.changeset/safeguard-propose-guard.md

Lines changed: 0 additions & 5 deletions
This file was deleted.

.changeset/safeguard-signal-flagger.md

Lines changed: 0 additions & 9 deletions
This file was deleted.

.changeset/safeguard-string-patterns.md

Lines changed: 0 additions & 5 deletions
This file was deleted.

packages/bash-trim/CHANGELOG.md

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,11 @@
11
# pi-bash-trim
22

3+
## 1.0.2
4+
5+
### Patch Changes
6+
7+
- ff4bf12: Only apply line deduplication when row trimming is actually needed. Output that fits within the token budget is no longer deduped.
8+
39
## 1.0.1
410

511
### Patch Changes

packages/bash-trim/package.json

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
{
22
"name": "pi-bash-trim",
3-
"version": "1.0.1",
3+
"version": "1.0.2",
44
"description": "Smart bash output trimming for pi to fit LLM context budgets",
55
"author": "mgabor3141",
66
"license": "MIT",

packages/safeguard/CHANGELOG.md

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,21 @@
11
# pi-safeguard
22

3+
## 2.0.0
4+
5+
### Major Changes
6+
7+
- af8ec18: Replace pattern matching with signal-based flagging. The flagger is now a wide-net boolean gate (high recall, no reasoning); the judge sees raw actions only — no flagger bias.
8+
9+
Broadened sensitive file detection beyond `.env*`: files outside cwd, dotfiles in `$HOME`, system paths, paths with secret keywords.
10+
11+
New signals: `rm -r`/`rm -f` flagged independently, inline code interpreters (`eval`, `bash -c`, `python -c`, `node -e`), `chmod u+s`/`g+s`, `su`/`doas`/`pkexec`, private key material, known API key formats.
12+
13+
### Minor Changes
14+
15+
- 04d1a1a: Add user-configurable `commands`, `patterns`, and `instructions` to safeguard config. Support project-level config at `.pi/extensions/pi-safeguard.json` (additive only — cannot weaken global settings). Add `\bsafeguard\b` to built-in string patterns.
16+
- f624ef6: Add `propose_trust` tool that lets the agent request permission when blocked by the security guardrail. The user sees the proposed trust rule with the agent's reasoning and can accept or reject with one keypress. Accepted rules work like `/guard` — they persist for the session.
17+
- e7b8ba5: Add string pattern matching in addition to AST-based detection — dangerous keywords like `sudo` are now caught anywhere in tool input text, not just as parsed command names. Fix post-denial circumvention check cascade.
18+
319
## 1.0.1
420

521
### Patch Changes

packages/safeguard/package.json

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
{
22
"name": "pi-safeguard",
3-
"version": "1.0.1",
3+
"version": "2.0.0",
44
"description": "LLM-as-judge guardrail for pi",
55
"author": "mgabor3141",
66
"license": "MIT",

0 commit comments

Comments
 (0)