The Agent Skills spec only defines name, description, license,
compatibility, metadata, and allowed-tools as valid frontmatter fields.
The trigger: /graphify line was non-spec, silently ignored by spec-
following hosts, and flagged by agentskills validate CI checks.
- gen.py: removed trigger emission from _render_frontmatter; added
_is_trigger_line() helper for roundtrip allow-list
- fragments/core/aider.md: removed hardcoded trigger: /graphify
- platforms.toml: removed trigger doc comment and trigger="" entries
- test_skillgen.py: replaced trigger-assertion tests with a single
test asserting no host has trigger: in frontmatter
- Regenerated all 125 skill artifacts
Routing intent is preserved: the description field already contains
"treated as a graphify query first" and "graphify-out/ exists".
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
graphify codebuddy install was writing CODEBUDDY.md and settings.json
but not copying the SKILL.md. Added _copy_skill_file("codebuddy") call
to match the --platform codebuddy path. README hook description updated
from "Glob and Grep" to "Bash search and file reads" to match actual
hook matchers (Bash + Read|Glob).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Three-part fix:
dedup.py: Pass 1 exact-merge now skips nodes with an empty source_file.
Previously all no-source_file nodes with the same label landed in one
bucket and were merged, destroying distinct symbols (third-party deps,
standalone functions) that happened to share a short name.
update.md (skillgen + all 13 host variants): the --update merge now
passes both deleted AND changed files to prune_sources, mirroring what
watch._rebuild_code already does correctly. Old nodes for re-extracted
files are pruned before fresh AST is inserted — no fuzzy reconciliation
needed, no cross-file collapse possible.
export.py: anti-shrink guard message now names fuzzy dedup as a
possible cause (not only "missing chunk files"), and advises a full
rebuild as the safe recovery path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
#1174: affected.py load_graph now forces directed=True before
node_link_graph, matching the identical fix in serve.py and __main__.py.
Undirected graphs (directed:false in graph.json) were causing in_edges
to fall back to a direction-blind scan, missing true callers and
reporting false positives. Regression test added.
#1173: post-commit and post-checkout hook bodies now read
graphify-out/.graphify_root before calling _rebuild_code, falling back
to Path('.') if absent. A scoped build (graphify src/) no longer gets
silently expanded to the full repo on the next commit. Tests added.
#1172: Step 9 cleanup split into rm -f for fixed files and
find -maxdepth 1 -delete for the chunk glob. Under fish/zsh an
unmatched glob aborts the entire rm -f line, leaving temp files on disk.
Fixed in the three skillgen source fragments and regenerated.
#1163: detect_incremental type guard on stored mtime — if the manifest
contains a dict-valued mtime (schema drift from older versions), coerce
to None rather than propagating a non-numeric into comparisons.
Regression test added.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
#1170 — replace nohup with cross-platform Python detach in git hooks.
Git for Windows MSYS has no nohup so post-commit/post-checkout hooks
silently failed. Now uses subprocess.Popen with DETACHED_PROCESS |
CREATE_NEW_PROCESS_GROUP on Windows, start_new_session=True on POSIX.
Quoting-safe (argv list). Fixes#1161.
#1169 — fix _is_sensitive false positives on topic-mentioning filenames.
token-economics-of-recall.md and password-policy-discussion.md were
silently dropped as secrets. Generic keywords (token/secret/password)
now only fire when the keyword ends the filename stem or the stem is
≤2 words. Specific patterns (.env/.pem/id_rsa etc.) remain unconditional.
#1165 — fix multi-word endpoint resolution in _score_nodes.
graphify path "AuthService" "UserRepo" never fired the exact-match bonus
because per-token comparison never equalled the full label. Now joins
normalized tokens and compares against the full label and its tokenized
form. O(1) per node, affects query_graph and shortest_path uniformly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
#1154: scope numpy>=2.0 constraint to python_version>='3.13' only.
numpy 1.26.4 ships no cp313 wheel so uv sync falls back to a source
build requiring a C compiler. The marker avoids forcing numpy 2.x on
3.10-3.12 users who have working 1.x environments.
#1160: codex platform skill now installs to .codex/skills/graphify/
instead of .agents/skills/graphify/. The hook already wrote to .codex/
so the skill destination was inconsistent. Propagates automatically
through install/uninstall (both read _PLATFORM_CONFIG dynamically).
Updated all codex-specific test assertions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
#1118 — prune stale AST nodes on full re-extraction (#1116)
Stamps every AST-extracted node with _origin="ast" in extract(). On a
full rebuild _rebuild_code drops any AST-marked node absent from the
fresh output even when its source file survives, fixing stale symbols.
Backward-compat: marker-less nodes from pre-1118 graphs survive one
cycle then self-heal.
#1110 — stop reading images and PDFs as garbage in headless extract
Images route through per-backend vision payloads (base64/data-URI/bytes
for claude/openai/bedrock); non-vision backends get _strip_pixels for
graceful degradation. PDFs reuse pypdf. 5MB cap, 20-image chunk limit.
#1159 — Salesforce Apex extractor (.cls, .trigger)
Regex-based extractor: classes, interfaces, enums, methods, triggers,
SOQL/DML edges. No new dependency. Dispatched as .cls and .trigger.
#1107 — Azure OpenAI Service backend (--backend azure)
Uses AzureOpenAI SDK client (from existing openai package). Auto-detects
when AZURE_OPENAI_API_KEY + AZURE_OPENAI_ENDPOINT both set. Uses
max_completion_tokens (not deprecated max_tokens).
#1103 — live PostgreSQL introspection (--postgres DSN)
graphify extract --postgres "postgresql://..." introspects tables, views,
routines, and FK relations via information_schema (SERIALIZABLE READ ONLY).
Credentials sanitized on error. New graphify[postgres] extra (psycopg3).
Union-resolved llm.py conflict: Azure functions + bedrock images= param.
Fixed test_image_vision.py mock to accept timeout= kwarg (our #1112).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
python -m graphify.serve graph.json --transport http --port 8080 serves
the same MCP tools over the Streamable HTTP transport (spec 2025-03-26)
so a single shared process can serve the graph for a whole team.
- _build_server() refactors server registration into a shared factory
(stdio behavior is byte-for-byte unchanged — all 52 existing tests pass)
- _ApiKeyMiddleware: raw ASGI (not BaseHTTPMiddleware) preserves SSE
streaming; constant-time compare; RFC-6750 case-insensitive Bearer;
blank-key normalized to no-auth
- DNS-rebinding protection via TransportSecuritySettings; wildcard binds
disable it and print an exposure warning when no api-key is set
- session_idle_timeout reaps idle stateful sessions (default 3600s) so a
long-running shared server does not leak memory on client disconnect
- Dockerfile + .dockerignore for containerized team deployment
- 16 new tests via in-process ASGI test client (importorskip-guarded)
- stdio remains the default; no change for existing setups
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
#1147 — builtin annotation nodes inflate god-node rankings:
Add _PYTHON_ANNOTATION_NOISE frozenset (str/int/bool/float/bytes/
MagicMock/Mock/AsyncMock/...) and apply it alongside
_PYTHON_TYPE_CONTAINERS in the annotation walker so scalar builtins
and test mocks are never created as nodes or emitted as edges.
Defense-in-depth guard in analyze.god_nodes filters _BUILTIN_NOISE_LABELS
so pre-existing graphs are also protected.
#1146 — package-form imports create disconnected islands:
from pkg import submod is now resolved to a file-level imports_from
edge when submod.py or submod/__init__.py exists on disk. Fix lives
in _collect_python_symbol_resolution_facts: when target resolves to
a __init__.py, each imported name is checked as a potential submodule
file and stored in _SymbolResolutionFacts.module_imports. Applied in
_apply_symbol_resolution_facts using stem-based canonical IDs.
#1145 — AST vs semantic node ID ghost duplicates:
build_from_json now runs a two-pass merge after adding all nodes:
collect AST nodes (source_location set) then find semantic ghosts
(same basename+label, no source_location). Ghosts are removed and
their IDs added to norm_to_id so all edges re-point to the AST node.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
TTY-only (suppressed in CI/pipes). Windows ANSI enabled via ctypes.
Wrapped in try/except so it can never crash the install.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend resolution now defers until after file detection. A code-only
corpus (pure AST, zero LLM calls) runs without any API key.
Key validation only fires when needs_llm=True (semantic_files non-empty
or --dedup-llm passed). Error message now names why a key is needed and
notes that code-only corpora need none.
Applied from PR #1123 with one fix: _clear_backend_keys in tests now
also clears AWS_PROFILE/REGION, OLLAMA_BASE_URL to prevent CI flakes
on machines with ambient credentials.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
_kiro_install was using a bare write_text that bypassed _copy_skill_file,
so the references/ sidecar and .graphify_version stamp were never written
despite kiro declaring skill_refs: "kiro". This left SKILL.md with 8 dead
references/*.md pointers on every install.
Fix: call _copy_skill_file("kiro", project=True) for the skill+sidecar+stamp,
keep the steering-file block inline. Uninstall now calls _remove_skill_file
which also cleans .graphify_version and references/.
Adds regression test asserting SKILL.md + references/ + .graphify_version +
no references.tmp + steering file, and that uninstall removes all of them.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
pytest.importorskip at module level matches the pattern used for other
optional extras (sql, dm) so CI passes without the optional dep.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds lower-bound version constraints matching tested versions in uv.lock. networkx>=3.4 is the binding constraint (edges= kwarg on node_link_data introduced in 3.4). rapidfuzz>=3.0 guards against the 2.x→3.x API break in rapidfuzz.distance. datasketch>=1.6 is conservative but correct. Also fixes stale uv.lock version metadata (0.8.30→0.8.31).
Fixes#777. Relativize manifest keys, .graphify_root, and cache source_file fields on persist; re-anchor on load. In-memory callers still see absolute paths. Symlink round-trip fixed in follow-up commit 8f09326.
Per adversarial review of the #1127 fix:
- Apply the filesystem-path allowlist to sys.executable before embedding it in
the generated hook script. Paths with metacharacters outside [a-zA-Z0-9/_.@:\-]
(spaces, dollar signs, backticks, semicolons) are replaced with an empty string
so the pinned probe is safely skipped rather than injecting shell commands that
execute on every git commit.
- Change _PINNED='...' to single-quote assignment so no shell expansion can occur
even if a character slipped through the allowlist (belt and suspenders).
- Quote $GRAPHIFY_PYTHON in both nohup exec lines so paths with spaces work
(common in Windows C:\Program Files\... installs).
- Update test to assert the sanitized value, not raw sys.executable, so the test
stays correct after any future sanitization changes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The issue reporter correctly noted that the hooks never consulted
graphify-out/.graphify_python -- the runtime interpreter source the skill and
README already rely on. Added as a second probe (after the pinned sys.executable,
before the launcher shebang and python3/python fallbacks). This covers Windows/Git
Bash where the uv-tool launcher is a binary with no parseable shebang, and where
the pinned exe may go stale after a reinstall.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
graphify hook install generated scripts that resolved the Python interpreter
purely at git-trigger time via 'command -v graphify'. GUI git clients (VS Code,
GitKraken), CI runners, and non-login shells often run with a minimal PATH that
omits ~/.local/bin -- the uv-tool / pipx launcher location. command -v graphify
returned empty, the python3/python fallbacks could not import graphify (it lives
only in the isolated venv), and the hook silently exited 0 with no output.
Fix: embed sys.executable of the currently-running install process as a pinned
first probe. Since 'graphify hook install' itself runs under the correct isolated
interpreter, sys.executable is always the right path. It is validated at
hook-runtime via 'import graphify' before use, so a stale pinned path safely
falls through to the existing dynamic detection rather than breaking the hook.
Also make the final fallback loud: print a diagnostic to stderr before exiting 0
so the failure is visible rather than invisible.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
test_manifest_written_after_extract assumes a docs corpus fails with no LLM
backend, but _run inherited the developer's environment. With a real
ANTHROPIC_API_KEY exported and the anthropic package present (now pulled by
[all]), the claude backend was detected and the extract succeeded, breaking the
'no backend' assertion. Strip the backend-selecting env vars in _run so the
tests hold regardless of the developer's shell. CI has no keys set, so it was
green there already.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
uv tool install graphifyy runs graphify in an isolated venv, so when a backend
needs anthropic/openai/boto3 the old 'Run: pip install anthropic' advice never
reached it. Worse, claude was the only backend with no [extra] at all, so a user
with ANTHROPIC_API_KEY set could not satisfy it without --with anthropic.
Add an anthropic optional extra (and include it in [all]), and replace the
package-missing ImportError messages with a shared _backend_pkg_hint that points
at 'uv tool install graphifyy[<extra>] --force' first, then the pip/venv path.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The Bash search hook only nudges grep/rg/find, so an agent that answers a
codebase question by Read-ing many source files one by one (the most common way
the graph gets skipped) slips right past it (#1114). Add _READ_SETTINGS_HOOK
matching Read|Glob: it fires only when graphify-out/graph.json exists, only for
a source/doc file outside graphify-out/, injects the same query-first
additionalContext, and never blocks (every branch fails open). Install and
uninstall now register and dedup both hooks idempotently.
Implemented independently rather than merging the community PR #1120; same idea,
our own hook (Read/Glob only, no fragile multi-file cat/head/tail heuristic).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
graphify install --project --platform antigravity went through the skill-only
branch (grouped with copilot/pi/kimi), so it copied SKILL.md but never wrote
.agents/rules/graphify.md or .agents/workflows/graphify.md - antigravity's
always-on layer - even though the project uninstall path removes them. Extract
the frontmatter + rules + workflows writes into _antigravity_finalize and call
it from both the global antigravity install and the project-scoped path.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Locks the body<->refs coupling: gemini ships claude's lean skill.md body but
resolves references through a separate path, so a real install with the real
claude bundle must leave every references/ pointer in SKILL.md resolvable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
F2: replace the header-trust ratio check with an authoritative bounded
streaming-decompression pass. The zip central-directory sizes are
attacker-controlled, so a member that under-declares its size could dodge the
declared-size checks; now every member is stream-decompressed with a hard byte
ceiling, so actual expansion past the cap is caught regardless of the headers.
The cheap declared-size pre-filter stays as a fast reject for honest bombs.
F3: _validate_ollama_base_url now resolves the host, so an alias that points at
a link-local/metadata IP is blocked too, not just literal IPs. The extract
command gains an early gate that turns the metadata block into a clean
'error: ...' exit(2) instead of a deep traceback; a warn toggle keeps the
single user-facing LAN warning in the in-flow call.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
F3: OLLAMA_BASE_URL pointing at a link-local or cloud-metadata address
(169.254.x, metadata.google.*, 0.0.0.0) now fails closed instead of only
warning, since no real Ollama host lives there and it is a classic SSRF
target. General LAN hosts still warn-and-allow.
F5: the Fortran C-preprocessor call now passes an absolute path. cpp has no
'--' end-of-options terminator, so an attacker-named corpus file like
'-I/etc/x.F90' could otherwise be parsed as a cpp option; an absolute path
always begins with '/' and cannot be.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Safi