When --doc-map-trusted-ref is provided, the --doc-map value is used as a
VCS API path parameter, not a local filesystem path. The early call to
validateWorkspacePath (which requires the file to exist locally) blocked
the trusted-ref code path when the doc-map did not exist in the local
checkout — defeating the feature's purpose in sparse checkouts or when
the file is only on the default branch.
Fix: guard the early validation with `&& *docMapTrustedRef == ""`.
Also fixes:
- review/docmap.go: correct ParseDocMapConfigContent godoc example to
match actual source format "owner/repo@ref:path"
- cmd/review-bot/main_test.go: add TestMainSubprocess_DocMapTrustedRefSkipsLocalValidation
to prevent regression
The doc-map YAML config was previously read from the local workspace
(the PR branch checkout). A malicious PR author could modify
.review-bot/doc-map.yml to map any path glob to sensitive design docs,
causing review-bot to fetch and inject those docs into the LLM prompt.
Fix: add --doc-map-trusted-ref (DOC_MAP_TRUSTED_REF) flag. When set to
a trusted ref (e.g. 'main'), the doc-map config is fetched from the VCS
API at that ref instead of from local workspace. A 404 from VCS is a
hard error (no silent fallback to local copy).
When unset, the local workspace is used with a security warning in the
logs pointing operators to the new flag.
Changes:
- review/docmap.go: add ParseDocMapConfigContent + parseDocMapBytes
helper to parse from in-memory content (fetched via VCS API)
- cmd/review-bot/main.go: add --doc-map-trusted-ref flag; Step 6c
branches on trusted-ref to fetch vs local-workspace load
- .gitea/actions/review/action.yml: add doc-map-trusted-ref input
- README.md: document new input
- CHANGELOG.md: security and feature entries
Tests:
- TestParseDocMapConfigContent_Valid/Empty/InvalidYAML/UnknownKeys
in review/docmap_test.go
Coverage: 53.0% cmd/review-bot
Finding #1 [MAJOR]: replace os.Stat with os.Lstat in checkStaleDocs to
prevent symlink traversal. Symlinks under repoRoot could probe arbitrary
host file existence; Lstat never follows them. Symlinked docs are now
treated as stale.
Finding #2 [MINOR]: resolve --repo-root with filepath.Abs +
filepath.EvalSymlinks before passing to checkStaleDocs, so a symlinked
repo-root cannot bypass the filepath.Rel escape guard.
Finding #3 [NIT]: reject backslashes in ValidateDocPath to prevent
Windows platform edge cases where a path separator may be normalised
differently by the host OS or VCS backend.
Tests added:
- TestCheckStaleDocs_SymlinkOutside: symlink inside repo → outside
- TestCheckStaleDocs_SymlinkInsideRepo: intra-repo symlink also rejected
- TestRunValidateDocmap_SymlinkRepoRoot: symlinked --repo-root resolves OK
- TestValidateDocPath_Backslash: backslash paths rejected
- Backslash cases added to TestValidateDocPath invalid slice
All go test ./... pass, go vet ./... clean.
Export review.ValidateDocPath and use it in checkStaleDocs before
calling os.Stat. Add filepath.Clean + filepath.Rel confinement check
as defense-in-depth to ensure doc paths from PR-controlled YAML
cannot probe filesystem locations outside repoRoot.
Also add tests covering: ../../etc/passwd, /etc/passwd, ../outside,
a valid present path, and a valid missing path.
Addresses security finding from security-review-bot on PR #142.
Adds FileCoveredByDocMap(cfg *DocMapConfig, file string) bool — a thin wrapper
over the existing unexported mappingMatches that lets cmd/ check per-file docmap
coverage without duplicating glob logic.
Also adds unit tests covering matched globs, non-matching paths, empty file,
and empty config.
- New --doc-map flag (DOC_MAP_FILE env var): path to YAML config mapping
source path globs to governing design docs
- New --doc-map-max-bytes flag (DOC_MAP_MAX_BYTES env var): cap on total
injected doc content, default 100KB
- review/docmap.go: DocMapConfig parsing, glob matching with ** support,
doc loading via VCS with directory expansion and size guard
- budget.Sections: new DesignDocs field, trimmed after conventions
- budget.buildResult: injects DesignDocs under ## Design Documents heading
- action.yml: doc-map and doc-map-max-bytes inputs wired to env vars
- CHANGELOG.md: created with unreleased entry
- Tests: ParseDocMapConfig, MatchDocs, globMatch, LoadMatchingDocs
- Clarify depth-aware short-circuit comment to unambiguously describe
the relationship between current depth and previous validation depth
- Add comment to MappingValueNode case explaining intentional depth+2
behavior from parent MappingNode perspective
- Restructure unmarshalYAMLWithDepthLimit doc comment as bullet list
covering all three safety checks (depth, multi-doc, strict fields)
- Replace t.Error with t.Fatal in TestYAMLEmptyFileRejection to remove
redundant nil guard on subsequent err.Error() call
- Add explicit case for *ast.MergeKeyNode in checkYAMLDepth switch to
make it clear this is an intentional leaf (no children to recurse)
rather than relying on the default case. Prevents future library
changes from silently bypassing depth checks.
- Add MaxPersonaFileSize bound check at the top of ParsePersonaBytes.
While callers already check size, the public API should defend itself
(defense in depth) against arbitrarily large inputs that could cause
excessive memory/CPU before AST validation runs.
- Add tests for both behaviors.
Addresses review #2879 findings.
- Add safety note on Strict() decoder not expanding aliases recursively,
since alias resolution uses the pre-validated AST (finding #1)
- Document that ast.Node map keys rely on pointer identity, which holds
because all goccy/go-yaml AST types are pointer receivers (finding #2)
- Clarify AnchorNode comment: effective depth budget is reduced for
anchor+alias pairs, not literally halved (finding #3)
- Improve test depth trace comment for accuracy (finding #4)
- Add HTML comment in CONVENTIONS.md referencing #91 for the two-step
process deviation (finding #5)
- Fix validated map comment: says 'minimum depth' but stores the maximum
depth at which a node was validated (overwritten on deeper visits).
- Replace dec.More() with explicit dec.Decode check for trailing JSON
content. More() is documented for use inside arrays/objects; the
explicit EOF check is clearer at the top-level stream.
MINOR fixes:
- docs/DESIGN-57-yaml-persona.md: fix Error Cases table entry to reflect
custom AST walk (checkYAMLDepth) instead of stale library-level reference
- review/persona.go: add EOF check after JSON decode to reject trailing
garbage after a valid JSON object (prevents silent acceptance of malformed
input like '{"name":"x"}garbage')
- review/persona_test.go: add TestJSONTrailingContentRejected test
NIT fixes:
- review/persona.go: add default case to checkYAMLDepth switch with
explanatory comment about scalar leaf nodes
- review/persona.go: document AnchorNode depth+1 conservative asymmetry
- review/persona.go: simplify redundant if-guard in ListBuiltinPersonas
- Move nodeCount increment after cycle detection to avoid over-counting
cyclic references (sonnet #2)
- Use underscores in test case names used as filenames (sonnet #3)
- Fix function comment: 'prevent silent data loss' → 'prevent confusing
behavior where additional documents are silently ignored' (sonnet #4)
- Mark design doc pseudocode as historical since implementation uses
goccy/go-yaml ast.Node, not gopkg.in/yaml.v3 yaml.Node (sonnet #5)
The global 'seen' set allowed anchored subtrees validated at a shallow
depth to be skipped when later referenced via alias at a greater depth.
This could let effective nesting exceed MaxYAMLDepth, enabling DoS.
Fix: replace the single 'seen' set with two tracking maps:
- validated (node -> min depth): only short-circuits when current depth
<= previously validated depth; re-checks at deeper contexts.
- visiting (node -> bool): per-path recursion stack for true cycle
detection (breaks alias loops without suppressing depth checks).
Add TestYAMLAliasDepthBypass that constructs a document with an
anchored 15-level subtree referenced via alias under 6 levels of
nesting, verifying the combined effective depth (22) is rejected.
Addresses security-review-bot findings on review #2774.
Fixes#87.
PR #58 incorrectly added gopkg.in/yaml.v3 (abandoned library) instead of
github.com/goccy/go-yaml as required by issue #57.
Changes:
- Replace gopkg.in/yaml.v3 with github.com/goccy/go-yaml v1.19.2
- Update review/persona.go to use goccy/go-yaml API:
- parser.ParseBytes for AST-based depth/node count checking
- yaml.Strict() decoder option instead of KnownFields(true)
- ast.Node types instead of yaml.Node for tree walking
- Update review/persona_test.go to use ast types for cycle tests
- Remove gopkg.in/yaml.v3 from go.mod and go.sum
All existing YAML tests pass with the new library.
MAJOR:
- LoadRepoPersonas: add MaxPersonaFileSize check before parsing to
prevent resource exhaustion from oversized YAML files committed
to target repositories
MINOR:
- isNotFoundError: tighten substring match to 'HTTP 404' only to
avoid masking auth/transport errors containing generic 'not found'
- main.go: remove duplicate flag.Parse() call
- main.go: add comment explaining nil map indexing is safe in Go
when LoadRepoPersonas returns an error
Tests updated to reflect the intentional behavior change in
isNotFoundError and added test case for oversized file rejection.
Implements #60.
- Add ParsePersonaBytes() for parsing personas from byte data
- Add LoadRepoPersonas() to fetch personas from repo via Gitea API
- Add MergePersonas() to combine built-in and repo personas
- Add GetBuiltinPersonasMap() helper
- Update main.go to load repo personas first, fall back to built-in
- Add giteaClientAdapter to bridge gitea.Client to review.GiteaClient
When --persona is specified, the bot now:
1. Attempts to fetch personas from .review-bot/personas/*.yaml
2. If the named persona exists in the repo, uses it
3. Otherwise falls back to built-in personas
This allows repos to define domain-specific personas (e.g., trading
experts for gargoyle, crypto experts for kms-lite) without modifying
the review-bot codebase.
1. Remove dead JSON fallback in LoadBuiltinPersona
- The embed directive only includes *.yaml files
- JSON fallback code could never succeed
- Simplified function to only try YAML
2. JSON parsing now rejects unknown fields
- Switched from json.Unmarshal to json.Decoder
- DisallowUnknownFields() matches YAML's KnownFields(true)
- Added test coverage for JSON unknown field rejection
3. Documented symlink support in LoadPersona
- os.Stat follows symlinks, so symlinks to regular files work
- Added doc comment explaining the behavior
- Added test for symlink support
Address security review findings:
MAJOR: Add cycle detection to checkYAMLDepth using a visited set
(seen map[*yaml.Node]struct{}) to prevent infinite recursion from
crafted YAML with self-referential aliases.
MINOR fixes:
- Add MaxYAMLNodes (1000) limit as defense-in-depth against
wide-but-shallow structures that bypass depth limits
- Increment depth when following alias targets (was incorrectly
passing same depth, allowing alias chains to bypass depth limit)
- Reject multi-document YAML files instead of silently ignoring
additional documents (prevents confusing silent data loss)
Tests added:
- TestYAMLAliasCycleDetection: Direct test of cycle detection logic
- TestYAMLMultiDocumentRejection: Verifies multi-doc files rejected
- TestYAMLNodeCountLimit: Verifies wide structures are rejected
- TestCheckYAMLDepthCycleDetectionDirect: Unit test with artificial cycle
Addresses PR #58 MINOR finding: YAML decoder now rejects unknown fields.
- Enable KnownFields(true) on YAML decoder to catch typos like
'focuss' or 'identiy' in persona files
- Since yaml.Node.Decode() doesn't support KnownFields, we now
do a two-pass decode: first pass checks depth limits, second
pass decodes with strict field checking
- Add tests for unknown field rejection at top-level and nested levels
MAJOR fixes:
- Remove false security claim about gopkg.in/yaml.v3 having built-in depth protection
- Add explicit YAML depth limiting via yaml.Node API (MaxYAMLDepth=20)
- Add file size limit for persona files (MaxPersonaFileSize=64KB)
- Add test for deeply nested YAML rejection
MINOR fixes:
- Add sort.Strings to ListBuiltinPersonas for deterministic ordering
- Update design doc to reflect actual library used (gopkg.in/yaml.v3)
- Update README: 'Zero dependencies' → 'Minimal dependencies'
- Add test for file size limit
- Add test for sorted persona list
- Add gopkg.in/yaml.v3 dependency (approved in CONVENTIONS.md)
- Update parsePersona to detect format by file extension
- Support both .yaml and .yml extensions (case-insensitive)
- Convert built-in personas to YAML format
- Add comprehensive tests for YAML parsing
- Update README with YAML examples and documentation
YAML provides cleaner multi-line strings via literal block scalars
and supports comments, making persona definitions more readable.
JSON remains supported for backwards compatibility.
Closes#57
MAJOR fixes:
- Remove external YAML dependency (github.com/goccy/go-yaml)
Per project convention: Go standard library only, zero dependencies.
Convert all persona files from YAML to JSON format.
- Fix TestValidateWorkspacePath error expectation
Go 1.21+ filepath.Join normalizes absolute paths differently.
MINOR fixes:
- Remove custom contains helper in persona_test.go (use strings.Contains)
- Add Unicode-safe CapitalizeFirst function for header titles
- ListBuiltinPersonas returns empty slice instead of nil on error
- Fix test comment about filepath.Join behavior
Documentation:
- Update README to reflect JSON-only persona format
- Update design doc with note about JSON decision
- Fix action.yml description for persona-file input
Add persona system for specialized review roles. Each persona defines:
- A specific review focus (security, architecture, documentation)
- Custom system prompt additions
- Personality/tone adjustments
Built-in personas: security, architect, docs
Custom personas: load from JSON via persona-file flag
Includes workspace validation to prevent path traversal attacks.
Closes#51
Addresses intermittent 'unexpected end of JSON input' failures where the
LLM response body is truncated in transit between the proxy and client.
Root cause: network-level truncation where io.ReadAll returns partial data
(observed in 3/50 CI runs through HAI proxy). The response body reading
was already using io.ReadAll correctly, but transient network issues
between the proxy and client can still cause partial reads.
Changes:
- Add Content-Length validation in doRequest: detect when fewer bytes
arrive than the server declared, triggering a retry
- Add retry logic in Complete: retries once on retryable errors (body
read failures, content-length mismatches) with a 500ms backoff
- Add parse-level retry in main: if ParseResponse fails, re-requests
from the LLM once before giving up (defensive, since retries always
succeed per issue evidence)
- Improve ParseResponse error diagnostics: log raw vs cleaned lengths
and a preview of the cleaned content to aid future debugging
Does NOT retry on API errors (4xx/5xx) or structural issues — only
transient body read problems.
Closes#47
When reviewer-name is set, prepend "# Security Review" / "# Sonnet Review"
etc. as a top-level header. Makes it immediately obvious which role each
review represents in the Gitea UI, especially when multiple reviews come
from the same bot account.
Sentinel-based cleanup:
- Reviews embed <!-- review-bot:NAME --> in body (hidden HTML comment)
- Cleanup matches by sentinel, not token identity
- Each reviewer-name is a logical identity (sonnet, gpt, security)
- Same token can run multiple review types without conflict
- No extra API scopes needed
System prompt file (--system-prompt-file / SYSTEM_PROMPT_FILE):
- Loads a local file with additional review instructions
- Appended to system base as "Additional Review Instructions"
- Enables specialized reviews (security, performance, etc.)
- Partially addresses #5
Security review:
- SECURITY_REVIEW.md prompt focused on vulnerabilities
- 3rd CI matrix entry using same token, different prompt
- Focus: injection, auth, secrets, input validation, crypto, races
CI changes:
- REVIEWER_NAME passed from matrix.name
- SYSTEM_PROMPT_FILE passed from matrix (empty for standard reviews)
- 3 reviewers: sonnet (general), gpt (general), security (focused)
- Account for truncation marker tokens when computing diff budget
(prevents EstTokens exceeding model limit in edge cases)
- Rune-safe truncation for both UserMeta and Diff (no split multi-byte)
- Fix misleading comment (1000 chars → ~1000 tokens/4000 chars)
- Extract marker strings as constants
- Add unit tests for BuildSystemBase and BuildUserMeta
Adds a budget package that estimates token usage and progressively
trims context to fit within model-specific limits.
Trim order (least important first):
1. Language patterns
2. Repository conventions
3. Full file context
4. Diff (truncated as last resort)
When content is trimmed, a note is appended to the user prompt so
the LLM knows context was reduced.
- New budget package with Fit(), EstimateTokens(), LimitForModel()
- Model limit table (GPT-4.1: 128K, GPT-5: 200K, Claude: 200K)
- Refactored review/prompt.go: BuildSystemBase() and BuildUserMeta()
extract non-trimmable content; old functions delegate to new ones
- main.go uses budget.Fit() instead of direct prompt assembly
- 7 unit tests covering all trim paths
Closes#19
Major improvements to review quality:
1. Full file context: fetch complete content of all modified files from
the PR branch and include as reference. This eliminates false-positive
"missing import" findings since the model sees the entire file.
2. Patterns repo: new --patterns-repo / PATTERNS_REPO flag fetches
language idiom files from a separate Gitea repo (e.g. rodin/elixir-patterns)
and includes them as review criteria.
3. Multi-file patterns: --patterns-files / PATTERNS_FILES accepts
comma-separated file paths to fetch from the patterns repo.
New API methods:
- GetPullRequestFiles: list changed files in a PR
- GetFileContentRef: fetch file content from a specific branch/ref
Prompt changes:
- BuildSystemPrompt now accepts (conventions, patterns)
- BuildUserPrompt now accepts fileContext parameter
- File context displayed before diff for model reference
- Patterns presented as "review criteria" in system prompt
Composite action updated with patterns-repo and patterns-files inputs.
The LLM was treating the diff as complete file context and flagging
"missing imports" for symbols that exist in unchanged portions of
the file. Added explicit instructions to the system prompt:
- Clarify that the diff is partial; unchanged code already exists
- Explicitly instruct: do not flag missing imports/types unless the
diff introduces a new usage without a corresponding addition
- Add rule: never flag undefined symbols that could exist in the
unchanged portions
- CLI binary with flag/env var configuration
- Gitea API client (PR metadata, diff, CI status, post review)
- OpenAI-compatible LLM client
- Structured review prompt with conventions support
- JSON response parser with validation
- Markdown review formatter for Gitea
- CI failure auto-detection (REQUEST_CHANGES)
- Dry-run mode for testing
Rodin
claw
Rodin
Aaron Weiker
rodin