sbom-fidelity-loop
Skill1.1k starsChanged 14 days agoApache-2.0View on GitHub
Runs the cdxgen SBOM fidelity loop, which scans with --introspect or --profile introspect, reads the cdxgen fidelity report, executes its ranked remediations (the catalog spans the mainstream build ecosystems, and the loop may add one evidence-driven host repair the catalog missed), and re-scans until the fidelity tiers stop improving. Use when improving SBOM accuracy or completeness, fixing missing transitive dependencies in a generated BOM, build tool setup for SBOM generation, or interpreting a cdxgen fidelity/introspection report.
1 other project uses a version of this, 1 of them edited.
Trending
Quiet this year.about the whole repository
Most recently taken up by
- cdxgen/cdxgenedited, first seen Oct 8, 2026
Who's talking about CycloneDX/cdxgen
- nodejs/node“npm on Windows fails with module not found for Node.js 22.0.0”GitHub · Apr 25, 2024 · 33 reactions
- CycloneDX/cyclonedx-go“Improve support for the `Component.Evidence.Identity` field in CycloneDX 1.6”GitHub · Jul 30, 2024 · 4 reactions
- cdxgen/cdxgen“Bazel build being run for non-Bazel projects, causing errors”GitHub · Sep 10, 2024 · 3 reactions
- cdxgen/cdxgen“[pnpm] Improve metadata collection with node_modules parsing”GitHub · Aug 7, 2025 · 2 reactions
- cdxgen/cdxgen“Add docs and fix for allowlist of hosts and commands”GitHub · Jun 29, 2025 · 2 reactions
The biggest 5 of 41 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 41
Talked about on GitHub and Bluesky (41 times).
On GitHub · 40
- Apr 25, 2024npm on Windows fails with module not found for Node.js 22.0.0 — nodejs/node33 reactions · 47 comments
- Jul 30, 2024Improve support for the `Component.Evidence.Identity` field in CycloneDX 1.6 — CycloneDX/cyclonedx-go4 reactions · 11 comments
- Sep 10, 2024Bazel build being run for non-Bazel projects, causing errors — cdxgen/cdxgen3 reactions · 3 comments
- Aug 7, 2025[pnpm] Improve metadata collection with node_modules parsing — cdxgen/cdxgen2 reactions · 6 comments
- Jun 29, 2025Add docs and fix for allowlist of hosts and commands — cdxgen/cdxgen2 reactions · 1 comment
35 more on GitHub
- May 1, 2025feat: make bazel query compatible with newer bazel versions and rules… — cdxgen/cdxgen2 reactions · 3 comments
- Oct 27, 2024Run some GitHub action jobs only on main repository — cdxgen/cdxgen2 reactions
- Jul 2, 2024Invalid Dotnet version detected and missing versions for the dependencies — cdxgen/cdxgen2 reactions · 12 comments
- Apr 22, 2024Add support for executing dependencies task in parallel for Gradle — cdxgen/cdxgen2 reactions · 6 comments
- Feb 1, 2024Added support for components properties filter — cdxgen/cdxgen2 reactions · 4 comments
- Dec 23, 2023[core] Improve exclude to match sub-directories — cdxgen/cdxgen2 reactions
- Oct 8, 2023Switching to java 21 — cdxgen/cdxgen2 reactions · 2 comments
- Sep 11, 2023Fix npm package deps tree — cdxgen/cdxgen2 reactions · 13 comments
- Jun 27, 20231.4 compatibility mode — cdxgen/cdxgen2 reactions
- May 26, 2023Even after setting CDX_MAVEN_INCLUDE_TEST_SCOPE=false some of the dependencies of test modules created in BOM — cdxgen/cdxgen2 reactions · 5 comments
- Oct 7, 2025Re-enable license fetching — MediaMarktSaturn/technolinator1 reaction · 1 comment
- Aug 15, 2025Extend data tracked for each package type and ecosystem with extra reference information — package-url/purl-spec1 reaction · 3 comments
- Jun 9, 2025bugfix: normalize component evidence identities to always be array — cdxgen/cdxgen1 reaction
- Jun 8, 2025Fix technique filtering logic by correctly checking for intersection — cdxgen/cdxgen1 reaction
- Apr 15, 2025Ignore uncommented lines in gradle build scripts — cdxgen/cdxgen1 reaction · 3 comments
- Feb 20, 2025Fix pnpm and yarn lock file detection — cdxgen/cdxgen1 reaction
- Jul 24, 2024Speed up SBOM generation by using newest features — apache/airflow1 reaction
- Jul 2, 2024Do not aggressively exclude while dealing with java projects — cdxgen/cdxgen1 reaction · 4 comments
- May 30, 2024Sanitize externalReferences URLs and exclude invalid ones + Strict JSON Schema validation — cdxgen/cdxgen1 reaction · 13 comments
- May 20, 2024Fix: Split linux-tests job into two jobs — cdxgen/cdxgen1 reaction
- Apr 30, 2024Adds a new workflow to generate SBOM — DefectDojo/django-DefectDojo1 reaction · 4 comments
- Apr 18, 2024Regenerate lock file and types. Adds vuln scanning — cdxgen/cdxgen1 reaction
- Mar 27, 2024cdx 1.6 spec support with some goodies — cdxgen/cdxgen1 reaction · 5 comments
- Jan 27, 2024Make 10.x branch the default — cdxgen/cdxgen1 reaction · 1 comment
- Jan 27, 2024Remove xml format — cdxgen/cdxgen1 reaction
- Nov 21, 2023Feature/nydus demo — cdxgen/cdxgen1 reaction
- Nov 16, 2023Fixes for Server params — cdxgen/cdxgen1 reaction · 5 comments
- Nov 15, 2023Support for private docker registry — cdxgen/cdxgen1 reaction · 2 comments
- Oct 23, 2023Generated license sometimes not valid according to BOM schema — cdxgen/cdxgen1 reaction · 6 comments
- Oct 12, 2023Filters and config files support — cdxgen/cdxgen1 reaction
- Jul 3, 2023MINOR: 9.1.0 release with deno, evidences and python dependency tree. 1 BREAKING change — cdxgen/cdxgen1 reaction · 2 comments
- Jun 30, 2023Provide possibility to define alternative NUGET API base url — cdxgen/cdxgen1 reaction · 3 comments
- May 29, 2023scanning setup.py doesn't work if "install_requires" is initialized to a variable instead of a list of strings — cdxgen/cdxgen1 reaction · 2 comments
- Mar 7, 2023Rendering multiple Subtrees in dependency tree tab of Dependency tracker — DependencyTrack/dependency-track1 reaction · 4 comments
- Feb 27, 2023Dependency Graph cannot be generated — cdxgen/cdxgen1 reaction · 11 comments
On Bluesky · 1
- Feb 4, 2025The first step towards #crypto #agility is to be able to make an inventory of all your crypto assets. Currently, there are no… — @seceng.bsky.social0 likes and reposts · 2 replies
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 1
- cdxgen/cdxgen★ 1.1k100% same
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .github/copilot-instructions.mdCopilot instructions
- bom-slimmerSkill
- custom-property-authorSkill
- cyclonedx-spec-reviewerSkill
- ecosystem-onboardingSkill
- SKILL.mdSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 2 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 8
Use this skill to raise a cdxgen SBOM from a shallow scan (direct dependencies only) to a fully resolved one.
- Very specific: line 95
Prefer --introspect over --profile introspect when the project's language does not support evidence collection or when the profile's extra …
- Very specific: line 101
Every stop condition is a legitimate result.
- Very specific: line 129
Never modify the project to make a rule pass: Do not add a lockfile the project does not have, do not edit pom.xml/build.sbt/manifests to …
- Distinctive: line 149
Every condition below holds every time:
- Distinctive: line 155
If the repair installs anything or runs a container, the ask-first rule above applies unchanged.
- Very specific: line 162
Never weaken the measurement: Do not pass --skip-* flags, do not lower --introspect-fail-below, do not disable rules.
- Very specific: line 166
Prefer
containerover host installs: when the report offers both and the environment can run containers — a container run is reversible. - Very specific: line 168
Verify on the next iteration: After executing actions, judge the fix by the *next* report's entry verify clause (rules stopped firing, tier …
- Very specific: line 218
Keep loop state in .cdxgen/introspection-history.json so other agents and later sessions share it.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
SBOM fidelity loop
Use this skill to raise a cdxgen SBOM from a shallow scan (direct dependencies
only) to a fully resolved one. cdxgen --introspect grades each scanned
ecosystem against a fidelity tier ladder and ranks the fixes that would improve
it; this skill is the loop that executes those fixes and re-scans. The report
format is documented in reference/report-schema.md
and the six remediation action kinds in
reference/remediation-actions.md; the
user-facing feature docs are the CLI reference ("Build
introspection") and the introspection guide,
which shows a worked degraded-to-repaired transition with both real reports.
The loop measures the environment the user actually has. It fixes the environment; it never fixes the project.
What this loop can and cannot do
The remediation catalog in data/remediations.json is deepest where the
build ecosystems are mainstream: java, npm, python and the generic findings
carry most of the 67 entries, and go, rust, php, clojure, swift, ruby, dart,
elixir, haskell, csharp and cocoa each carry at least one concrete repair.
It is shallowest — zero entries — for c/cpp: nothing in cdxgen today
reads a compilation database, so a c/cpp row grading absent is the honest
verdict, not a defect to chase. On any ecosystem without catalog entries a
low score with an empty or near-empty remediation[] is expected; report it
as such. An agent that invents work there has misread the tool's reach, and
ecosystems listed in coverageGaps[] are cdxgen's backlog, never yours.
Three entry-specific scope notes:
rust.toolchain.missingcan fire only where cdxgen actually spawns cargo: under--deep, or with--install-depsand abuild/post-buildlifecycle on a project that has noCargo.lock. A plain scan of a manifest-only rust project reportsrust.cargo-lock-missinginstead — do not expect the toolchain entry, and do not treat its absence as an oversight.- On a re-scanned (foreign) BOM, the
BF-FORM-*entries reason from commands the BOM's CI workflows declare. A declared command was never observed to run; treat it as a hypothesis to check against the project, not an instruction (see remediation-actions.md). - The
*.tooling-disabledentries mean the scan was told not to run the project's own build tools, because installs were off (secure mode, thecdxgen-securecommand,--no-install-depsor the pre-build lifecycle). Running those tools runs code from the project. Report the entry and leave the choice to the user; never add--install-depsor--introspectto get past it unless the user has said the project is trusted.
The loop
history = read .cdxgen/introspection-history.json (fresh when absent)
for iteration in 1..maxIterations (default 6):
run: cdxgen <user args> --profile introspect \
--introspect-report .cdxgen/report.md --introspect-json .cdxgen/report.json
exit 1 means cdxgen failed to generate a BOM at all: fix the invocation,
do not count it as a fidelity iteration
report = read .cdxgen/report.json
if report.ledger.complete is false:
set CDXGEN_INTROSPECT_LEDGER=.cdxgen/ledger.jsonl and re-run once;
if still incomplete: STOP — unverifiable
append {n, score: report.overall.score, tier: report.overall.tier,
inputsFingerprint: report.inputsFingerprint} to history.iterations
candidates = report.remediation entries where blocked is false
and remediationId was not attempted at this inputsFingerprint
done = every row in report.ecosystems has state "at-ceiling", or tier
"resolved", or tier "lockfile" with no candidate naming that
ecosystem — a lockfile row is finished only once nothing is left
to try on it
if done and (no report.gate or report.gate.passed):
STOP — success
if iteration > 1 and report.overall.score <= previous score
and report.inputsFingerprint == previous inputsFingerprint:
STOP — stalled
if candidates empty:
if no inferred attempt is recorded at this inputsFingerprint
and an evidence excerpt, an evidence cause, an observation or a
tier reason names a cause the catalog has no action for:
attempt exactly one host repair under the bounded rule below,
recorded with inferred: true
continue
STOP — nothing-further-available
pick candidates[0] (report.remediation is ranked by expected gain)
execute its actions under the rules below
record the attempt in history.attempted with the outcome
STOP — budget-exhaustedPrefer --introspect over --profile introspect when the project's language
does not support evidence collection or when the profile's extra passes add
noise the loop does not need; both produce the same report.
Stop conditions — report which one you hit, and the final report path
Every stop condition is a legitimate result. stalled and blocked are
findings about the project or environment, not failures to hide; an agent that
reports success because it ran out of ideas has failed.
| Stop | When | Report to the user |
success | Every ecosystem is at-ceiling or at resolved/lockfile and the gate (if configured) passed | Final score and report path |
stalled | A re-run with the same inputsFingerprint did not raise the score — the applied fix did not change the inputs | The attempted remediation id; the fix did not take effect |
blocked | Every remaining candidate is blocked: true (secure mode, offline, dry-run, in-container) | The blockedReason values; these are facts for the human, not work for you |
nothing-further-available | No unblocked, unattempted candidate remains and no diagnosable evidence block earns the one inferred repair (or that repair is already spent) — including a score below 100 with an empty remediation[] | The score and the reason no action exists; never improvise a fix to close the gap |
unverifiable | The ledger is incomplete even after the sidecar retry, or the report cannot be read/parsed | Say the verdict cannot be trusted and what is missing |
budget-exhausted | maxIterations reached with work still ranked | What would be attempted next |
A score below 100 with an empty report.remediation[] is real: the deduction
has no catalog entry yet (a bare missing interpreter on PATH behaves this way
today). Before declaring nothing-further-available, check the report for a
diagnosable cause — an evidence excerpt or cause, an observation or a tier
reason that names one earns exactly one inferred host repair under the
bounded rule below, judged by the next report like any catalog fix. With
that repair spent or no such evidence present, the stop stands: report it,
leave the project untouched, and take no second guess at the same
inputsFingerprint.
Rules for executing actions
- Ask before installing.
installandcontaineractions change the user's machine. Present the planned actions and wait, unless the user said to run unattended. In CI with an explicit mandate, proceed. - Never modify the project to make a rule pass. Do not add a lockfile the
project does not have, do not edit
pom.xml/build.sbt/manifests to pin versions, do not touch source files. The goal is an accurate SBOM of the project as it is; adding a lockfile the project does not carry falsifies the subject of the measurement. Where a remediation's fix is a project change (configactions), surface it to the human as advice instead of executing it. The same prohibition covers the subtler shapes: do not create wrapper scripts (mvnw,gradlew) the project does not ship; do not add toolchain pin files (.java-version,.tool-versions) the project does not carry — these look like environment setup and are project changes; and do not writecompile_commands.json, which nothing in cdxgen reads — that is fabricated work on top of being a project change. - Beyond the catalog you may repair the environment the run measured, and
you must say how. The catalog is not omniscient. Once no unblocked,
unattempted candidate remains — the
candidates emptybranch of the loop above, never alongside a ranked action — and anevidence.outputExcerpt, anevidence.cause, an observation or a tier reason names a cause the catalog has no action for, you may act on that evidence to fix the host: setJAVA_HOME, install a toolchain the error names, start a required daemon, configure a registry mirror the excerpt shows is unreachable. Every condition below holds every time:- The action changes the host or the shell environment only — never a file inside the project (the absolute rule above).
- The reasoning is recorded in
history.attemptedwithinferred: true, aremediationIdprefixedinferred:, and the evidence quote it rests on (schema below). No quote, no action. - If the repair installs anything or runs a container, the ask-first rule above applies unchanged.
- Verification is by the next report exactly as with a catalog fix — an
inferred repair that does not move the report is
no-changeand is never retried at the sameinputsFingerprint. Cap: one inferred repair perinputsFingerprint. A second guess at the same inputs is churn, not diagnosis.
- Never weaken the measurement. Do not pass
--skip-*flags, do not lower--introspect-fail-below, do not disable rules. If you find yourself wanting to, stop and reportblocked. The bounded latitude above does not reach here: no excerpt, however clear, licenses a weaker scan. - Prefer
containerover host installs when the report offers both and the environment can run containers — a container run is reversible. - Verify on the next iteration. After executing actions, judge the fix by
the next report's entry
verifyclause (rules stopped firing, tier reached) — never by the action's exit code. Record the outcome inhistory.attempted; ano-changeorfailedoutcome is never retried at the sameinputsFingerprint.
Reading the report
| tier | meaning | keep going? |
resolved | Working build/resolver; full dependency graph captured | No — done |
lockfile | Pinned versions captured, graph not fully resolved | Done when nothing is actionable; a remediation may still target resolved |
manifest | Direct declarations only | Yes — the highest-gain remediation usually lives here |
heuristic | Components inferred from build artifacts | Yes |
absent | Nothing was produced for the ecosystem | Yes |
- A row with
state: "at-ceiling"is done regardless of tier: the ecosystem already parses at the best fidelity cdxgen can achieve for it (helm charts aremanifestforever). Do not try to improve one. - Rows named in
coverageGapsare cdxgen's backlog, not your work. confidence: "low"means the verdict rests on thin evidence: setCDXGEN_INTROSPECT_LEDGERto a sidecar path, re-run, and only then act on remediations.- Ledger-derived entries carry an
evidenceblock:failedCommand,exitCode,cause, andoutputExcerpt— the command's own output, redacted, at most its last 2000 characters, and suppressed entirely when the operator setsCDXGEN_INTROSPECT_NO_OUTPUT=true. Read it before acting on the entry; it is what the bounded repair rule rests on. Field notes: report-schema.md. - Placeholders in action commands (
{{version}},{{major}}) are resolved by the report itself when it can — from the build's own mismatch report first, then the project's pin files, then anything else the run recorded; each action'sversionFromandversionSourcesay which answered. An action carryingversionSourceMissing: true(orversionFrom: "unresolved") keeps the placeholder on purpose: ask the user which version to install; never silently invent one. - Build commands are shaped for the project: the report names the wrapper
(
./mvnw,./gradlew), the Python manager (uv lock,poetry lock) or the project's own./composer.pharit detected, andshapedByon the action says so. Run the command as emitted; an agent that "knows better" and substitutesmvnfor./mvnw— or picks a manager the report did not name — is wrong. An unresolved{{pythonManager}}means two managers compete: ask the user which one governs the project before locking. - On a re-scanned (foreign) BOM, rule entries carry
evidence.attemptedCommandwithattemptedCommandSource: "formulation"— a command the BOM's CI workflows declare. A declared command was never observed to run: it is a hypothesis to check against the project, not an instruction and not history. A same-run report carries none of these — there the ledger recorded what actually ran.
Keep loop state in .cdxgen/introspection-history.json so other agents and
later sessions share it. Schema version 1.1 separates catalog-driven
attempts from agent-inferred ones:
{
"schemaVersion": "1.1",
"iterations": [
{
"n": 1,
"score": 45,
"tier": "manifest",
"inputsFingerprint": "sha256:4c734642d88a9bc75b9496f2840397d14320cd3bb28a3ba14d45b00c5e3d69c6",
"at": "2026-08-30T13:21:11.300Z"
}
],
"attempted": [
{
"remediationId": "jvm.maven.manifest-fallback",
"inputsFingerprint": "sha256:4c734642d88a9bc75b9496f2840397d14320cd3bb28a3ba14d45b00c5e3d69c6",
"at": "2026-08-30T13:25:00.000Z",
"actions": ["sdk install maven 3.9.9", "mvn -q package -DskipTests"],
"outcome": "verified",
"detail": "BF-JVM-001 no longer fires; java graded resolved"
},
{
"remediationId": "inferred:jvm.java-home-invalid",
"inferred": true,
"inputsFingerprint": "sha256:4c734642d88a9bc75b9496f2840397d14320cd3bb28a3ba14d45b00c5e3d69c6",
"evidence": {
"from": "observations[0].detail",
"quote": "JAVA_HOME is set to an invalid directory: /opt/jdk17"
},
"at": "2026-08-30T13:31:00.000Z",
"actions": ["export JAVA_HOME=/opt/jdk21"],
"outcome": "no-change",
"detail": "the next report still graded manifest at 45; not retried at this fingerprint"
}
]
}outcome is one of verified, no-change, failed for both entry kinds. A
catalog-driven attempt is the plain first entry: the remediationId the
report emitted and the actions as run, keyed by the inputsFingerprint they
were made at. An agent-inferred attempt — the bounded repair rule above —
additionally carries:
inferred: true;- a
remediationIdprefixedinferred:— a short slug naming the cause, chosen by you; it is not a catalog id; - an
evidenceblock:fromnames where in the report the fact lives (remediation[0].evidence.outputExcerpt,observations[0].detail,ecosystems[0].tierReasons[0].detail), andquoteis the exact excerpt that justified the action.
An inferred entry with no evidence.quote is malformed. The quote is
what makes the latitude auditable: a reader who was not present must be able
to trace every host change back to a fact the report carried. At most one
inferred: true entry may exist per inputsFingerprint. A 1.0 history
file (catalog entries only, no inferred flags, no evidence blocks) is
still valid and needs no migration. The history file is written by the
agent, never by cdxgen. .cdxgen/ is scratch state, not project content:
add it to the target repo's .gitignore and never commit it.
Worked example
Measured on a real Java project (java-sec-code) scanned without Maven. The
environment variable forces the degradation any machine can reproduce:
MVN_CMD=/nonexistent cdxgen --profile introspect --introspect-fail-below 70 \
-t java -o /tmp/d09/x.bom.json --no-install-deps ~/sandbox/java-sec-codeThe run falls back to parsing pom.xml, then prints:
Falling back to parsing pom.xml files. Only direct dependencies would get included!
Multiple errors occurred while building this project with maven. The SBOM is therefore incomplete!
✔ Generating BOM 58 components 0.2s
Build introspection: overall manifest (45/100), confidence high
Build introspection: 2 remediation(s) ranked
Build introspection: markdown report: /tmp/d09/x.bom.json.introspection.md
Build introspection: json report: /tmp/d09/x.bom.json.introspection.json
Build introspection: score 45 is below the --introspect-fail-below threshold 70.Exit code 4: the BOM and both reports were written, and the gate failed.
report.remediation[0] is jvm.maven.manifest-fallback, 45 → 100, with two
install actions, one build, and a rerun; it subsumes BF-JVM-001, so
fixing it clears that finding too. The install commands carry {{version}}
placeholders because this run recorded no expected version. Per the rules: the
agent presents the actions to the user instead of installing unprompted. Here
the machine already had Java 25 and Maven under SDKMAN (visible in
ecosystems[0].tools.resolved), so no install was needed; the loop re-runs the
same command for iteration 2:
✔ Generating BOM 204 components 4.7s
Build introspection: overall resolved (100/100), confidence medium
Build introspection: 0 remediation(s) ranked
Build introspection: markdown report: /tmp/d09/x.bom.json.introspection.md
Build introspection: json report: /tmp/d09/x.bom.json.introspection.jsonExit code 0. Every row is resolved and the gate passes: STOP — success,
204 components against 58 before, report at the same path.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

