@mokuyoaxis/agent-guard 0.2.3-rc2 → 0.2.4-hotfix.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (82) hide show
  1. package/README.md +132 -84
  2. package/README.zh-CN.md +109 -75
  3. package/adapters/INTEGRATION.md +4 -4
  4. package/adapters/claude/README.md +3 -3
  5. package/adapters/claude/harness/README.md +1 -1
  6. package/adapters/codex/harness/README.md +2 -2
  7. package/adapters/dsh/README.md +24 -15
  8. package/adapters/dsh/harness/README.md +75 -19
  9. package/adapters/dsh/harness/bootstrap.patch.yml +5 -0
  10. package/adapters/dsh/harness/bootstrap_init.mjs +48 -0
  11. package/adapters/dsh/harness/guard_lab.py +194 -32
  12. package/adapters/dsh/harness/native_session.py +41 -20
  13. package/adapters/dsh/harness/profiles.py +74 -22
  14. package/adapters/dsh/lib/index.js +1 -1
  15. package/adapters/kimi-code/README.md +15 -14
  16. package/adapters/kimi-code/harness/README.md +3 -3
  17. package/adapters/kimi-code/harness/guard_lab.py +1 -1
  18. package/core/__init__.py +1 -1
  19. package/core/classifier.py +66 -17
  20. package/core/lab.py +24 -1
  21. package/core/policy.py +28 -13
  22. package/core/recovery.py +256 -23
  23. package/core/redaction.py +157 -55
  24. package/core/trash_index.py +193 -0
  25. package/docs/README.md +104 -0
  26. package/docs/{architecture.md → design/architecture.md} +31 -6
  27. package/docs/{compatibility.md → design/compatibility.md} +80 -4
  28. package/docs/{threat-model.md → design/threat-model.md} +4 -4
  29. package/docs/{harness-capabilities.md → guides/harness-capabilities.md} +9 -9
  30. package/docs/guides/trash-index.md +106 -0
  31. package/docs/{development-note-unguarded-deletion.md → history/development-note-unguarded-deletion.md} +8 -9
  32. package/docs/{friction.md → history/friction.md} +129 -4
  33. package/docs/{secret-guard-analysis.md → history/secret-guard-analysis.md} +23 -22
  34. package/docs/{secret-guard-architecture.md → history/secret-guard-architecture.md} +6 -4
  35. package/docs/{secret-guard-design.md → history/secret-guard-design.md} +19 -15
  36. package/docs/lab/cases/injection-case-fake-ai-api.md +346 -0
  37. package/docs/{guard-lab.md → lab/guard-lab.md} +18 -11
  38. package/docs/lab/tasks/guard-lab-read-trial-task.md +7 -0
  39. package/docs/{release-notes-0.2.0-rc1.md → releases/release-notes-0.2.0-rc1.md} +5 -5
  40. package/docs/{release-notes-0.2.0.md → releases/release-notes-0.2.0.md} +7 -7
  41. package/docs/{release-notes-0.2.1.md → releases/release-notes-0.2.1.md} +6 -6
  42. package/docs/{release-notes-0.2.2.md → releases/release-notes-0.2.2.md} +9 -10
  43. package/docs/{release-notes-0.2.3-rc1.md → releases/release-notes-0.2.3-rc1.md} +4 -2
  44. package/docs/{release-notes-0.2.3-rc2.md → releases/release-notes-0.2.3-rc2.md} +9 -9
  45. package/docs/releases/release-notes-0.2.3.md +84 -0
  46. package/docs/releases/release-notes-0.2.4-hotfix.1.md +75 -0
  47. package/docs/releases/release-notes-0.2.4.md +89 -0
  48. package/docs/{test-report-claude-code-harness.md → reports/test-report-claude-code-harness.md} +5 -5
  49. package/docs/{test-report-codex-gpt-5.6-sol.md → reports/test-report-codex-gpt-5.6-sol.md} +10 -11
  50. package/docs/reports/test-report-dsh-0.2.4-ptc.md +79 -0
  51. package/docs/{test-report-dsh-guard-lab.md → reports/test-report-dsh-guard-lab.md} +1 -1
  52. package/docs/{test-report-dsh-native-session.md → reports/test-report-dsh-native-session.md} +1 -1
  53. package/docs/{test-report-dsh-paired-lab.md → reports/test-report-dsh-paired-lab.md} +1 -1
  54. package/docs/{test-report-dsh-real-followup.md → reports/test-report-dsh-real-followup.md} +1 -1
  55. package/docs/{test-report-dsh-retry-evidence.md → reports/test-report-dsh-retry-evidence.md} +1 -1
  56. package/docs/{test-report-dsh-v0.1.1.md → reports/test-report-dsh-v0.1.1.md} +4 -4
  57. package/docs/{test-report-kimi-code-block.md → reports/test-report-kimi-code-block.md} +22 -23
  58. package/docs/{test-report-kimi-guard-lab.md → reports/test-report-kimi-guard-lab.md} +3 -3
  59. package/docs/{test-report-release-readiness-0.2.3.md → reports/test-report-release-readiness-0.2.3.md} +79 -1
  60. package/docs/reports/test-report-workbuddy-windows-core.md +53 -0
  61. package/docs/{test-report-zcode-glm-flash.md → reports/test-report-zcode-glm-flash.md} +11 -11
  62. package/package.json +1 -1
  63. package/skills/delete-guard/SKILL.md +40 -3
  64. package/skills/delete-guard/references/policy.md +29 -9
  65. package/skills/delete-guard/scripts/check.py +19 -0
  66. package/skills/delete-guard/scripts/restore.py +18 -6
  67. package/skills/delete-guard/scripts/status.py +52 -0
  68. package/skills/exfil-guard/SKILL.md +37 -7
  69. package/skills/exfil-guard/references/channels.md +1 -1
  70. package/skills/exfil-guard/references/rules.md +84 -16
  71. package/skills/exfil-guard/scripts/_payload.py +99 -0
  72. package/skills/exfil-guard/scripts/check_span.py +14 -81
  73. package/skills/exfil-guard/scripts/sanitize.py +113 -34
  74. package/docs/release-notes-0.2.3.md +0 -45
  75. /package/docs/{host-drift.md → guides/host-drift.md} +0 -0
  76. /package/docs/{publishing-from-ephemeral-environments.md → guides/publishing-from-ephemeral-environments.md} +0 -0
  77. /package/docs/{guard-lab-clean-task.md → lab/tasks/guard-lab-clean-task.md} +0 -0
  78. /package/docs/{guard-lab-trial-task.md → lab/tasks/guard-lab-trial-task.md} +0 -0
  79. /package/docs/{test-report-codex-gpt-6-astra-high.md → reports/test-report-codex-gpt-6-astra-high.md} +0 -0
  80. /package/docs/{test-report-dsh-0.1.5-rc.1.md → reports/test-report-dsh-0.1.5-rc.1.md} +0 -0
  81. /package/docs/{test-report-dsh-0.2-upgrade.md → reports/test-report-dsh-0.2-upgrade.md} +0 -0
  82. /package/docs/{test-report-dsh-read-redaction.md → reports/test-report-dsh-read-redaction.md} +0 -0
@@ -31,7 +31,6 @@ import argparse
31
31
  import json
32
32
  import os
33
33
  import sys
34
- import time
35
34
 
36
35
  import _bootstrap # noqa: F401
37
36
 
@@ -40,65 +39,7 @@ from core.redaction import (
40
39
  CHANNELS, UNREACHABLE_CHANNELS, get_channel, scan_text,
41
40
  )
42
41
  from core import policy
43
-
44
-
45
- # A non-interactive caller that opens a pipe but never writes to it would
46
- # otherwise sit in `sys.stdin.read()` until the *host's* tool timeout kills
47
- # the call: the guard presents as a hang instead of a refusal, and a
48
- # mistyped invocation costs the caller a whole timeout window. Wait a
49
- # bounded time for the first byte - once a producer has spoken, read to EOF
50
- # as usual - and refuse loudly if nothing ever arrives. Interactive use is
51
- # unchanged. Override with AGENT_GUARD_STDIN_TIMEOUT (seconds).
52
- DEFAULT_STDIN_TIMEOUT_SEC = 10.0
53
-
54
-
55
- class StdinIdle(Exception):
56
- """No payload arrived on stdin within the caller's patience."""
57
-
58
-
59
- def _stdin_timeout() -> float:
60
- try:
61
- value = float(os.environ.get("AGENT_GUARD_STDIN_TIMEOUT", ""))
62
- except (TypeError, ValueError):
63
- return DEFAULT_STDIN_TIMEOUT_SEC
64
- return value if value > 0 else DEFAULT_STDIN_TIMEOUT_SEC
65
-
66
-
67
- def read_payload(stream=None, timeout=None) -> str:
68
- """Read the payload from stdin without hanging a non-interactive caller.
69
-
70
- A TTY is a human ending the stream with Ctrl-D: keep blocking. Anything
71
- else is a producer that is expected to have spoken already, so the wait
72
- for its first byte is bounded. A pipe that closes without data (EOF) is
73
- a legitimately empty payload and returns "" for the normal scan path.
74
- """
75
- stdin = stream if stream is not None else sys.stdin
76
- if stdin.isatty():
77
- sys.stderr.write("exfil-guard: reading payload from stdin "
78
- "(Ctrl-D to finish)\n")
79
- return stdin.read()
80
- try:
81
- import select
82
- except ImportError: # pragma: no cover - non-POSIX
83
- return stdin.read()
84
- budget = _stdin_timeout() if timeout is None else timeout
85
- deadline = time.monotonic() + budget
86
- while True:
87
- remaining = deadline - time.monotonic()
88
- if remaining <= 0:
89
- raise StdinIdle(
90
- f"no payload arrived on stdin within {budget:g}s; pipe the "
91
- "text in, or run it on a terminal to type it")
92
- try:
93
- ready, _, _ = select.select([stdin], [], [],
94
- min(remaining, 0.5))
95
- except (OSError, ValueError):
96
- # stdin is not selectable on this platform (e.g. a Windows
97
- # pipe). Fall back to the blocking read rather than refusing a
98
- # caller whose payload may be perfectly good.
99
- return stdin.read()
100
- if ready:
101
- return stdin.read()
42
+ from _payload import PayloadTooLarge, StdinIdle, read_payload, resolve_context
102
43
 
103
44
 
104
45
  def build_redaction_plan(spans, decisions, channel) -> list:
@@ -129,30 +70,11 @@ def main() -> int:
129
70
  "repo-local exemption file for that path")
130
71
  args = ap.parse_args()
131
72
 
132
- workspace = args.workspace
133
- if workspace is None:
134
- from core.classifier import discover_workspace
135
- workspace = discover_workspace(os.getcwd())
136
-
137
- mode = args.mode
138
- if mode is None:
139
- trash_root = os.environ.get(
140
- "AGENT_GUARD_TRASH", os.path.join(workspace, ".agent-trash"))
141
- try:
142
- mode = policy.load_mode(trash_root)["mode"]
143
- except Exception: # noqa: BLE001 - state is advisory
144
- mode = policy.MODE_NORMAL
73
+ workspace, mode = resolve_context(args.workspace, args.mode)
145
74
 
146
75
  channel_name = (args.channel or "").strip().lower()
147
76
  channel = get_channel(channel_name)
148
77
 
149
- try:
150
- text = read_payload()
151
- except StdinIdle as exc:
152
- return _fail(str(exc), args.as_json)
153
- except (OSError, UnicodeDecodeError) as exc:
154
- return _fail(f"could not read stdin: {exc}", args.as_json)
155
-
156
78
  out = {
157
79
  "channel": channel_name,
158
80
  "channel_known": channel is not None,
@@ -160,9 +82,20 @@ def main() -> int:
160
82
  "channel_persistence": channel.persistence if channel else None,
161
83
  "workspace": workspace,
162
84
  "mode": mode,
163
- "size_bytes": len(text.encode("utf-8", errors="replace")),
164
85
  }
165
86
 
87
+ try:
88
+ text = read_payload(max_bytes=args.max_bytes)
89
+ except PayloadTooLarge:
90
+ out.update(scanned=False, reason="payload exceeds scan cap")
91
+ return _emit(out, policy.DECISION_BLOCK,
92
+ policy.CODE_BLOCK_OUTPUT_UNSCANNABLE, [], [], args, 2)
93
+ except StdinIdle as exc:
94
+ return _fail(str(exc), args.as_json)
95
+ except (OSError, UnicodeError, ValueError):
96
+ return _fail("could not read a bounded UTF-8 payload", args.as_json)
97
+ out["size_bytes"] = len(text.encode("utf-8", errors="strict"))
98
+
166
99
  if channel is None:
167
100
  # An unknown channel name is a configuration defect, and a channel
168
101
  # the guard cannot name is one it cannot certify. Fail closed.
@@ -14,13 +14,15 @@ so a shell pipeline can use it without writing code:
14
14
  Two properties are load-bearing:
15
15
 
16
16
  * **Format preservation.** A redacted secret keeps the shape a reader needs
17
- to recognise it (`sk-proj-Ab…` -> `sk-proj-<REDACTED>`), never the bytes.
17
+ to recognise it (`sk-proj-Ab…` -> `sk-<REDACTED>`), never the bytes.
18
18
  For paths the placeholder is `<PATH>`.
19
19
  * **Leak freedom.** The plan carries offsets and placeholders only, and the
20
20
  output is built by *replacing* the matched ranges - so no code path here
21
21
  can echo a matched secret into stdout, the plan, or an error message.
22
22
 
23
- Exit codes: 0 applied (or nothing to do) | 1 error.
23
+ Exit codes: 0 allow/applied | 2 block | 3 ask | 1 error.
24
+ Refusals and errors emit no payload. External plans are revalidated against
25
+ the current scan; they are never an authorization to bypass its verdict.
24
26
  """
25
27
  from __future__ import annotations
26
28
 
@@ -34,6 +36,8 @@ import _bootstrap # noqa: F401
34
36
 
35
37
  from core import policy, redaction
36
38
  from core.redaction import get_channel, scan_text
39
+ from _payload import (PayloadTooLarge, StdinIdle, read_bounded_text,
40
+ read_payload, resolve_context)
37
41
 
38
42
  REDACTED = "<REDACTED>"
39
43
  PATH_PLACEHOLDER = "<PATH>"
@@ -46,12 +50,13 @@ def placeholder_for(span) -> str:
46
50
  """The replacement text for one span; derived from shape, never content."""
47
51
  if span.family != redaction.FAMILY_SECRET:
48
52
  return PATH_PLACEHOLDER
49
- if span.rule_id == redaction.RULE_PRIVATE_KEY_BLOCK:
53
+ if span.rule_id in (redaction.RULE_PRIVATE_KEY_BLOCK,
54
+ redaction.RULE_CONNECTION_PASSWORD):
50
55
  return REDACTED
51
56
  # Keep the vendor prefix: it is public information (it says *which*
52
57
  # credential kind leaked) and it is what makes the output reviewable.
53
58
  match = re.match(r"^(sk-|rk_|ghp_|gho_|ghu_|ghs_|ghr_|github_pat_|"
54
- r"glpat-|xox[baprs]-|AKIA|ASIA|AGPA|AIDA|AROA|ANPA|"
59
+ r"glpat-|xox[baprs]-|pypi-|AKIA|ASIA|"
55
60
  r"sk_live_|rk_live_)", span.raw)
56
61
  if match:
57
62
  return match.group(1) + REDACTED
@@ -70,26 +75,61 @@ def plan_for_span(span, channel=None) -> dict:
70
75
  }
71
76
 
72
77
 
73
- def apply_plan(text: str, plan, verify: bool = True) -> str:
74
- """Replace every planned range, back to front so offsets stay valid.
75
-
76
- Blocks whose span overlaps a gap (stale plan) are skipped rather than
77
- applied blindly: a wrong splice would silently corrupt the payload,
78
- which is the false-SANITIZE failure mode the analysis warns about.
79
- """
78
+ def _plan_pieces(text, plan):
79
+ """Validate every range before constructing any output."""
80
+ if not isinstance(plan, list):
81
+ raise ValueError("redaction plan must be a list")
80
82
  pieces = []
81
- for entry in sorted(plan, key=lambda e: e["start"], reverse=True):
83
+ for entry in plan:
84
+ if not isinstance(entry, dict) or not {"start", "end", "placeholder"}.issubset(entry):
85
+ raise ValueError("invalid redaction plan entry")
82
86
  start, end = entry["start"], entry["end"]
87
+ if type(start) is not int or type(end) is not int:
88
+ raise ValueError("redaction offsets must be integers")
83
89
  if start < 0 or end > len(text) or start >= end:
84
90
  raise ValueError("redaction plan has an out-of-range span")
91
+ if not isinstance(entry["placeholder"], str) or not entry["placeholder"]:
92
+ raise ValueError("invalid redaction placeholder")
85
93
  pieces.append((start, end, entry["placeholder"]))
86
- result = text
94
+ pieces.sort()
95
+ for previous, current in zip(pieces, pieces[1:]):
96
+ if current[0] < previous[1]:
97
+ raise ValueError("redaction plan has overlapping spans")
98
+ return pieces
99
+
100
+
101
+ def apply_plan(text: str, plan, verify: bool = True) -> str:
102
+ """Use validated original coordinates to assemble output in one pass.
103
+
104
+ `verify` is retained for callers. Structural validation is mandatory;
105
+ only `sanitize_text` can verify a plan against a fresh policy decision.
106
+ """
107
+ pieces = _plan_pieces(text, plan)
108
+ chunks = []
109
+ cursor = 0
87
110
  for start, end, placeholder in pieces:
88
- result = result[:start] + placeholder + result[end:]
89
- if verify and REDACTED in result and result.count(REDACTED) > len(pieces):
90
- # Another marker was already present: harmless, but worth knowing.
91
- pass
92
- return result
111
+ chunks.extend((text[cursor:start], placeholder))
112
+ cursor = end
113
+ chunks.append(text[cursor:])
114
+ return "".join(chunks)
115
+
116
+
117
+ def _checked_plan(text, plan, derived):
118
+ """Only a plan equivalent to the current scanner's plan may be replayed."""
119
+ # Structural validation is independent of policy/metadata checks.
120
+ _plan_pieces(text, plan)
121
+ expected = sorted(derived, key=lambda entry: entry["start"])
122
+ supplied = sorted(plan, key=lambda entry: entry["start"])
123
+ if len(supplied) != len(expected):
124
+ raise ValueError("redaction plan does not match the current scan")
125
+ for entry, actual in zip(supplied, expected):
126
+ if set(entry) - set(actual):
127
+ raise ValueError("invalid redaction plan fields")
128
+ if any(entry.get(key) != actual[key] for key in ("start", "end", "placeholder")):
129
+ raise ValueError("redaction plan does not match the current scan")
130
+ if any(value != actual[key] for key, value in entry.items()):
131
+ raise ValueError("redaction plan metadata does not match the current scan")
132
+ return expected
93
133
 
94
134
 
95
135
  def sanitize_text(text: str, channel_name: str, workspace=None,
@@ -97,26 +137,58 @@ def sanitize_text(text: str, channel_name: str, workspace=None,
97
137
  """Scan + decide + apply, in one call (the library form)."""
98
138
  channel = get_channel(channel_name)
99
139
  if channel is None:
100
- raise ValueError("unknown egress channel: " + str(channel_name))
101
- scan = scan_text(text, channel=channel_name, workspace=workspace)
140
+ raise ValueError("unknown egress channel")
141
+ workspace, mode = resolve_context(workspace)
142
+ exemption = redaction.load_exemption(workspace)
143
+ scan = scan_text(text, channel=channel_name, workspace=workspace,
144
+ exemption=exemption)
102
145
  if not scan.scanned:
103
- raise RuntimeError(scan.error)
146
+ raise RuntimeError("payload could not be scanned")
104
147
  ctx = policy.PolicyContext(workspace=workspace or "", trash_root="",
105
148
  base_dir=workspace or "",
106
- mode=policy.MODE_NORMAL)
149
+ mode=mode)
107
150
  verdicts = policy.decide_spans(scan.spans, ctx=ctx, channel=channel)
151
+ decision = (policy.worst(verdicts).decision if verdicts else
152
+ policy.DECISION_ALLOW)
108
153
  derived = [plan_for_span(span, channel)
109
154
  for span, verdict in zip(scan.spans, verdicts)
110
155
  if verdict.decision == policy.DECISION_SANITIZE]
111
- applied = plan if plan is not None else derived
156
+ if decision not in (policy.DECISION_ALLOW, policy.DECISION_SANITIZE):
157
+ return {"text": "", "plan": derived, "decision": decision}
158
+ applied = derived if plan is None else _checked_plan(text, plan, derived)
159
+ rewritten = apply_plan(text, applied)
160
+ checked = scan_text(rewritten, channel=channel_name, workspace=workspace,
161
+ exemption=exemption)
162
+ if not checked.scanned:
163
+ raise RuntimeError("sanitized payload could not be scanned")
164
+ remaining = policy.decide_spans(checked.spans, ctx=ctx, channel=channel)
165
+ if remaining and policy.worst(remaining).decision != policy.DECISION_ALLOW:
166
+ raise RuntimeError("sanitized payload did not pass verification")
112
167
  return {
113
- "text": apply_plan(text, applied),
168
+ "text": rewritten,
114
169
  "plan": applied,
115
- "decision": policy.worst(verdicts).decision if verdicts else
116
- policy.DECISION_ALLOW,
170
+ "decision": decision,
117
171
  }
118
172
 
119
173
 
174
+ def _unique_object(pairs):
175
+ obj = {}
176
+ for key, value in pairs:
177
+ if key in obj:
178
+ raise ValueError("duplicate plan field")
179
+ obj[key] = value
180
+ return obj
181
+
182
+
183
+ def _read_plan(filename):
184
+ with open(filename, "rb") as stream:
185
+ text = read_bounded_text(stream, redaction.max_scan_bytes())
186
+ obj = json.loads(text, object_pairs_hook=_unique_object)
187
+ if not isinstance(obj, dict) or not isinstance(obj.get("redaction_plan"), list):
188
+ raise ValueError("invalid plan document")
189
+ return obj["redaction_plan"]
190
+
191
+
120
192
  def main() -> int:
121
193
  ap = argparse.ArgumentParser(description=__doc__)
122
194
  ap.add_argument("--channel", default=os.environ.get(
@@ -129,17 +201,24 @@ def main() -> int:
129
201
  help="print the plan, not the sanitized text")
130
202
  args = ap.parse_args()
131
203
 
132
- workspace = args.workspace or os.environ.get("AGENT_GUARD_WORKSPACE")
133
- text = sys.stdin.read()
134
- plan = None
135
- if args.plan:
136
- with open(args.plan, "r", encoding="utf-8") as fh:
137
- plan = json.load(fh).get("redaction_plan", [])
138
204
  try:
205
+ workspace, _ = resolve_context(args.workspace)
206
+ text = read_payload()
207
+ plan = _read_plan(args.plan) if args.plan else None
139
208
  result = sanitize_text(text, args.channel, workspace, plan)
140
- except (ValueError, RuntimeError) as exc:
141
- print(f"sanitize: {exc}", file=sys.stderr)
209
+ except PayloadTooLarge:
210
+ print("sanitize: BLOCK [BLOCK_OUTPUT_UNSCANNABLE] input exceeds cap",
211
+ file=sys.stderr)
212
+ return 2
213
+ except StdinIdle:
214
+ print("sanitize: no payload arrived on stdin within the wait", file=sys.stderr)
215
+ return 1
216
+ except (OSError, UnicodeError, ValueError, RuntimeError):
217
+ print("sanitize: payload or plan validation failed", file=sys.stderr)
142
218
  return 1
219
+ if result["decision"] in (policy.DECISION_BLOCK, policy.DECISION_ASK):
220
+ print("sanitize: " + result["decision"] + " payload withheld", file=sys.stderr)
221
+ return 2 if result["decision"] == policy.DECISION_BLOCK else 3
143
222
  if args.dry_run:
144
223
  print(json.dumps(result["plan"], ensure_ascii=False, indent=2))
145
224
  return 0
@@ -1,45 +0,0 @@
1
- # Agent Guard 0.2.3 — release-note draft
2
-
3
- This is a draft for the planned stable release. The current source remains
4
- `0.2.3-rc2`; these notes do not mean `0.2.3` has been published. See the
5
- [readiness review](test-report-release-readiness-0.2.3.md) for outstanding gates.
6
-
7
- ## Changes
8
-
9
- - Windows/Core fail-closed fixes from rc1: known destructive vocabulary is
10
- checked across shell dialects, and Windows path normalization/recovery has
11
- focused real-Windows CI evidence. Arbitrary Windows harness E2E remains
12
- outside the claim.
13
- - `agent-guard-lab`: disposable synthetic fixtures, a bounded loopback
14
- observer, clean/positive/injection/snapshot controls, explicit output scans
15
- and privacy-minimal reports. Missing or unhealthy evidence is inconclusive.
16
- - Reviewed real-harness capture helpers and matched comparisons: an effective
17
- guard-off baseline is required before declaring mitigation. Historical
18
- evidence retains its original format and channel coverage.
19
- - DSH adapter translation for `0.2.0-rc.2` Shell handles, with the tested old
20
- default deletion path retained. Core Decision Protocol and deletion rules
21
- remain shared and independent of DSH.
22
- - Default-off DSH complete text-read protection on reviewed, artifact-pinned
23
- providers. It replaces structured values so native content and metadata
24
- regenerate; incomplete, unsupported or unscannable reads are withheld.
25
- - Bounded native session v4 / receipt v5 support alongside preserved v3
26
- interpretation, including retry-attempt evidence. Surface replay, image
27
- offload and arbitrary session shapes remain unsupported.
28
- - DSH/Kimi Lab startup failures retain the original observer error and never
29
- launch a host task or acquire a successful capture receipt.
30
-
31
- ## Evidence and limits
32
-
33
- The offline Lab controls establish instrument calibration. Historical bounded
34
- model trials and fresh no-model native checks are listed in the
35
- [rc2 notes](release-notes-0.2.3-rc2.md) and
36
- [current review](test-report-release-readiness-0.2.3.md). The retained new-host
37
- real-model clean capture remains inconclusive; no new injection L2 result is
38
- claimed. A refused Guard plugin does not stop all other host tools, and Web
39
- inventory alone does not establish activation in an agent's scope.
40
-
41
- No production dependency was added. Python Core remains Python 3.9+;
42
- the optional pinned DSH read feature uses Node 22 and the Kimi TOML capture
43
- helper requires Python 3.11+. Lab evidence is limited to declared channels;
44
- same-UID isolation, general host indexing/uploads and external exfiltration
45
- containment are outside its guarantee.
File without changes