Repository navigation
Expand file tree
/
Copy path.scan-secrets-allow
More file actions
202 lines (181 loc) · 10.5 KB
/
Copy path.scan-secrets-allow
File metadata and controls
202 lines (181 loc) · 10.5 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
# scan-secrets allowlist — paths/files where token-shape patterns are intentional
# documentation, test fixtures, planted-secret examples for the scanner test harness, etc.
# Format: one path glob per line. # for comments.
# Basename globs work (e.g. *.md matches any .md file). Full path globs also work.
# .env files: these are gitignored by design and hold real credentials locally.
# They must NEVER be committed; the scanner correctly skips them here because
# they are already covered by .gitignore. The pre-push hook gates committed content,
# not the local working tree .env files.
.env
.env.*
helix_code/.env
helix_code/.env.*
# Documentation files with example keys/tokens (not real secrets)
docs/COMPLETE_CLI_REFERENCE.md
docs/troubleshooting/guide.md
docs/superpowers/plans/*.md
docs/superpowers/specs/*.md
docs/improvements/05_phase_0_evidence.md
# RENDERED .html EXPORTS of the .md files above (HXC-331, 2026-08-13).
#
# Root cause, not a symptom: this allowlist had ZERO .html globs, while the
# §11.4.65 doc-export wave touched 411 .html + 393 .pdf siblings in
# 437a893f (2026-06-10) — of which 408 .html / 390 .pdf were NEWLY ADDED and
# 3 of each were updates (`git diff-tree -r --diff-filter=A --name-only`
# vs the unfiltered form; both counts are right, they answer different
# questions, so state which one is meant). The .md SOURCES were allowlisted 2026-07-11 — two of
# the four explicitly (lines above), two via the `docs/superpowers/plans/*.md`
# wildcard — but their exports never were, so the SAME six key-shaped strings
# were suppressed in the source and flagged in the render. Verified 1:1: with
# the allowlist active the four .md sources scan rc=0; with no allowlist
# loadable they produce the same six findings. So this is NOT an export-only
# artefact and the remedy is NOT "fix the generator".
#
# EXPLICIT paths, never a `plans/*.html`-style wildcard — measured on this
# tree: a wildcard covering the directory ALSO silences a freshly planted
# key-shaped string in a sibling file (rc=0, silent), while these explicit
# entries leave that plant caught (rc=1). A wildcard here would trade six
# known-benign findings for an unbounded future blind spot.
#
# Honest boundary (§11.4.6): each entry blinds a scanner to the WHOLE file for
# all future content, which is why the set is four named files and not a glob.
# The six suppressed findings were each classified individually
# (fabricated/illustrative fixtures, no real credential).
#
# WHICH scanner these four entries actually bind, measured (an earlier revision
# of this comment asserted "BOTH scanners ... skip the .pdf siblings as binary";
# that is right about the outcome and wrong about the mechanism, so it is
# restated here rather than repeated):
# * secret_scan.sh reads EVERY tracked file and decides per file. It is the
# scanner these four entries operatively bind. It skips the .pdf siblings by
# an explicit binary probe — `grep -Iq . -- <pdf>` -> rc=1 on all four — so
# it never reads them. That one IS a binary skip.
# * scan-secrets.sh never reaches EITHER extension in its real full-repo mode:
# it restricts the walk with a `--include=` extension list (scan-secrets.sh
# FILE_INCLUDES) that contains neither `.html` nor `.pdf`. That is an
# extension filter, NOT a binary skip — the script has no binary probe at
# all. Measured in that mode against fabricated plants: a `.md` plant is
# flagged (rc=1, positive control) while `.html` and `.pdf`-named plants are
# not. NOTE when re-testing: FILE_INCLUDES is applied ONLY when the scan
# target is `.`; passing an explicit directory takes the else-branch, which
# deliberately disables the filter for the test harness and WILL appear to
# scan .html/.pdf. Testing that branch and concluding "it scans .html" is a
# measurement artefact — it was made once already while writing this note.
#
# So "no finding on the .pdf siblings" is a statement about scanner REACH, not
# about the PDFs being clean. They are not clean: `pdftotext <pdf> | grep -cE -e
# '<key shapes>'` recovers 5 of the 6 suppressed strings across the four exports
# (the 6th only under a dash-tolerant variant — extraction mangles that file's
# dash runs; a 31,919-byte extraction confirms the reader itself worked). Use
# `grep -e` when probing: these patterns begin with `-----` and grep parses a
# bare pattern as a flag, which returns an ERROR that reads exactly like a 0.
# Anything that makes those streams greppable — an uncompressed re-export, a
# content-extracting scanner, a change to FILE_INCLUDES — surfaces them.
docs/improvements/05_phase_0_evidence.html
docs/troubleshooting/guide.html
docs/superpowers/plans/2026-05-04-phase-0-foundation-cleanup.html
docs/superpowers/plans/2026-05-06-p2-f22-aider-git-auto-commit.html
# HelixCode test files — contain fake/example API keys and JWTs as test fixtures
helix_code/internal/llm/*_test.go
helix_code/internal/worker/*_test.go
# Confirmed fabricated-fixture false positives against scripts/secret_scan.sh
# (§11.4.135/§11.4.138 pre-commit guard), 2026-07-11 remediation. Both are
# already covered by the *_test.go wildcards two lines up; listed explicitly
# here too so the specific confirmed-safe files are documented by name and
# stay covered even if a future edit narrows the wildcard scope.
# - vertexai_provider_test.go: fabricated GCP service-account JSON fixture
# (fake PEM body + type/private_key_id markers) exercising the VertexAI
# provider's credential-parsing path — not a real credential.
# - ssh_pool_test.go: fabricated RSA PEM private-key header fixture
# exercising the SSH worker pool's key-loading path — not a real key.
helix_code/internal/llm/vertexai_provider_test.go
helix_code/internal/worker/ssh_pool_test.go
# F22 autocommit secret-filter tests — intentional fake AKIA/sk-/xox*-/gh*_
# fixtures used to validate the redaction regex. The strings are fabricated
# placeholder bodies and would never authenticate against any real service.
helix_code/internal/autocommit/secret_filter_test.go
helix_code/internal/autocommit/committer_test.go
helix_code/internal/autocommit/summariser_test.go
helix_code/tests/integration/cmd/p2f22_challenge/main.go
helix_code/tests/e2e/test-bank/*_test.go
helix_code/tests/e2e/test-bank/performance_security_tests.go
helix_code/tests/e2e/test_bank/*
helix_code/tests/security/authentication_test.go
helix_code/tests/security/authorization_test.go
# HelixCode docs — JWT examples in API documentation (not real tokens)
helix_code/docs/general/API_DOCUMENTATION.md
# HelixCode API spec — JWT examples in OpenAPI documentation (not real tokens)
helix_code/api/openapi.yaml
# Top-level user manual — example/placeholder credentials in setup snippets
docs/user_manual/README.md
docs/user_manual/tutorials/*.md
# Challenges test/security files with example patterns
challenges/panoptic/tests/security/*_test.go
challenges/pkg/env/redact_test.go
challenges/pkg/logging/redacting_logger_test.go
challenges/panoptic/docs/SECURITY.md
# helix_qa test files — contain fake API keys for unit test validation
helix_qa/pkg/llm/google_test.go
# Security package test files — contain PEM stubs/examples for unit tests
security/pkg/securestorage/securestorage_test.go
# Example_Resources — third-party cookbook/examples directory, not production code
# (legacy name; renamed to cli_agents_resources/ in P1.5-WP2)
Example_Resources/*
cli_agents_resources/*
# cli_agents — third-party reference CLI agents (renamed from Example_Projects/ in P1.5-WP2).
# Each subdir is its own upstream repo; planted/test/sample credentials and JWTs
# in their fixtures and docs are NOT real secrets and are not under our governance.
cli_agents/*
# helix_qa — submodule with deeply nested third-party tooling (chroma, docling,
# llama-index, midscene, perfetto, signoz, skyvern, etc.). Their test fixtures
# and documentation contain dummy API keys, JWTs, and PEM-shaped tokens that are
# not real secrets. The meta-repo only ships a gitlink; their content is governed
# upstream.
helix_qa/tools/opensource/*
# SSH key test fixtures (ephemeral, gitignored, not committed) — old keys on disk
# (legacy kebab path; WP7 normalised to ssh_keys/)
helix_code/test/workers/ssh-keys/id_rsa
helix_code/test/workers/ssh_keys/id_rsa
helix_code/test/workers/ssh_keys/id_rsa.pub
# External/vendor package-lock files (integrity hashes, not secrets)
**/package-lock.json
package-lock.json
# Scanner infrastructure — this script and its test contain planted secrets intentionally
scripts/scan-secrets.sh
scripts/test-scan-secrets.sh
scripts/test-scan-secrets-range-perf.sh
.scan-secrets-allow
# panoptic security tests + documentation contain canonical AWS example access
# key (AKIAIOSFODNN7EXAMPLE — AWS-published fake), `sk-1234567890abcdef`-shape
# placeholder tokens, and PEM-banner fragments. These ARE the inputs the
# security tests assert get redacted; they are not real secrets. The
# subscriber project ships a gitlink; their content is governed upstream.
panoptic/tests/security/*
panoptic/docs/SECURITY.md
scripts/secret_scan.sh
scripts/secret_scan_test.sh
# CLI-agent config installer redaction tests + their reproduction record
# (§11.4.135/§11.4.138 pre-commit guard), 2026-09-07.
#
# WHY THESE ARE FALSE POSITIVES: test_agent_configs.sh builds agent config
# files containing fabricated credentials for the SOLE PURPOSE of asserting
# that install_agent_configs.sh masks them. The literals are the test's INPUT
# and its expected-redaction list — the same string appears twice, once as the
# planted fixture and once in the `for leaked in ...` assertion that proves it
# was scrubbed. Flagged shapes: sk- (OpenAI), AIzaSy (Google), and a generic
# PRIVATE KEY header. None would authenticate against any service.
#
# WHY THE ALLOWLIST RATHER THAN AN INLINE MARKER: the scanner accepts either.
# An inline REDACTED/EXAMPLE marker would require editing the credential
# literals themselves — but several sit inside heredoc'd JSON where a shell
# comment is not possible, and mangling the bodies would make the fixtures
# LESS faithful to the real key shapes the redaction regex must match. That is
# the "mangle content just to satisfy this scanner" outcome this file's own
# design notes argue against.
#
# HONEST BOUNDARY (§11.4.6), same as the four .html entries above: each entry
# blinds the scanner to the WHOLE file for all future content. Both are named
# EXPLICITLY, never a `scripts/tests/*` or `docs/qa/**` glob, which would trade
# these known-benign findings for an unbounded future blind spot.
scripts/tests/test_agent_configs.sh
docs/qa/agent_config_fanout_fixes_20260906T091122Z/REPRODUCTIONS.md