podtrace

https://github.com/gma1k/podtrace/blob/main/CONTRIBUTING.md

Podtrace uses the Developer Certificate of Origin (DCO), not a CLA or CAA.

Policy and explanation: https://github.com/gma1k/podtrace/blob/main/CONTRIBUTING.md#developer-certificate-of-origin
The section links to https://developercertificate.org/ and states what signing off means for this project: the contributor asserts they wrote the code or have the right to submit it under Apache-2.0, that they have employer permission where the employer owns the copyright, and that the name and email in the trailer are public and permanent. Contributors add the trailer with git commit -s.

Enforcement: https://github.com/gma1k/podtrace/blob/main/.github/workflows/dco.yml
The workflow runs on every pull request and verifies that each commit in the PR carries a Signed-off-by: trailer whose email matches the commit author's email. Missing sign-offs fail the check and block the merge. Machine-generated commits from bot accounts (renovate, dependabot, github-actions) are exempt, since a bot cannot make the DCO assertion.

Contributors are also reminded in the pull request template: https://github.com/gma1k/podtrace/blob/main/.github/PULL_REQUEST_TEMPLATE.md

Podtrace documents a maintainer-led governance model in GOVERNANCE.md:
https://github.com/gma1k/podtrace/blob/main/GOVERNANCE.md

Roles: the document defines Contributors (anyone opening an issue, answering in Discussions, improving docs, or sending a PR; no formal status required or conferred) and Maintainers (hold merge rights). Maintainers are named in https://github.com/gma1k/podtrace/blob/main/MAINTAINERS.md and their per-path review authority is encoded in https://github.com/gma1k/podtrace/blob/main/CODEOWNERS, which GitHub enforces as required approvals.

Decision-making: decisions are made in the open on the pull request or issue that proposes them, with the bar scaled to how hard the change is to reverse. Bug fixes, docs, tests, and dependency bumps need a PR with CODEOWNERS approval. New features, probes, and adapters need a feature request or epic issue first so the shape is agreed before code is written. Changes to the public surface (CRDs, CLI flags, Helm values, env vars) additionally require an explicit read of STABILITY.md, with breaking changes marked in the commit subject and accompanied by a migration note. Security fixes are handled privately under SECURITY.md until an advisory ships.

Dispute resolution: disagreements are resolved by discussion on the thread. Where consensus is not reached, maintainers decide, and are required to state their reasoning in the thread rather than merging silently.

Role changes: maintainership follows demonstrated work (a track record of merged non-trivial contributions, reviews that catch real problems, and sustained engagement) and is extended by invitation from existing maintainers. Maintainers who step back are recorded as emeritus rather than removed. Governance changes themselves go through the same PR process as code, proposed as their own PR.

Related: code of conduct and its enforcement path at https://github.com/gma1k/podtrace/blob/main/CODE_OF_CONDUCT.md, and support and communication channels at https://github.com/gma1k/podtrace/blob/main/SUPPORT.md

Podtrace has adopted the Contributor Covenant, version 2.1, as its code of conduct, posted at the standard repository-root location:
https://github.com/gma1k/podtrace/blob/main/CODE_OF_CONDUCT.md

Because the file is CODE_OF_CONDUCT.md at the root, GitHub detects it automatically and surfaces it in the repository's Community Standards profile and in the contributor-facing prompts on issues and pull requests. The canonical upstream text is linked at https://www.contributor-covenant.org/version/2/1/code_of_conduct/

Scope: the code of conduct applies to all project-managed spaces (issue tracker, pull requests, discussions, and official off-platform communication) and to individuals officially representing the project in public. It addresses behavior within the project community only.

Reporting: concerns are reported privately to the maintainer. All reports are reviewed and investigated promptly and fairly, and the maintainer is obligated to respect the privacy and security of the reporter.

Enforcement: the project applies the Contributor Covenant Enforcement Guidelines (https://www.contributor-covenant.org/version/2/1/code_of_conduct/#enforcement-guidelines), with four escalating levels: correction, warning, temporary ban, and permanent ban. Enforcement authority rests with the maintainer listed in https://github.com/gma1k/podtrace/blob/main/MAINTAINERS.md, consistent with the maintainer-led model documented in https://github.com/gma1k/podtrace/blob/main/GOVERNANCE.md

Podtrace documents its roles and their responsibilities across two linked files, plus a machine-enforced ownership map.

Roles are defined in https://github.com/gma1k/podtrace/blob/main/GOVERNANCE.md#roles
Contributors are anyone who opens an issue, answers a question in Discussions, improves docs, or sends a pull request; no formal status is required and none is conferred. Maintainers hold merge rights.

Maintainer responsibilities and required tasks are listed in https://github.com/gma1k/podtrace/blob/main/MAINTAINERS.md#what-maintainers-do
Maintainers: (1) review and merge pull requests and are the required approvers under CODEOWNERS; (2) triage incoming issues, apply area and type labels, and close duplicates; (3) cut releases following the release flow in CONTRIBUTING.md; (4) handle private vulnerability reports under SECURITY.md; (5) decide what lands in the public surface governed by STABILITY.md.

Several of these tasks carry documented, measurable commitments. SECURITY.md (https://github.com/gma1k/podtrace/blob/main/SECURITY.md) binds vulnerability handling to acknowledgement within 3 business days, triage and confirmation within 7 business days, and a default 90-day coordinated disclosure window. CODE_OF_CONDUCT.md (https://github.com/gma1k/podtrace/blob/main/CODE_OF_CONDUCT.md) assigns code-of-conduct enforcement to the maintainer under the Contributor Covenant Enforcement Guidelines.

Contributor-side responsibilities are documented in https://github.com/gma1k/podtrace/blob/main/CONTRIBUTING.md, covering DCO sign-off, Conventional Commit format, the required test plan, and regeneration of manifests and clientsets when API types change.

Who holds which role: MAINTAINERS.md names the current maintainer (Ghassan Malke, GitHub @gma1k) with contact details and areas of ownership. Per-path review authority is encoded in https://github.com/gma1k/podtrace/blob/main/CODEOWNERS, which GitHub enforces as required approval, making the role assignment machine-checkable rather than only asserted in prose. MAINTAINERS.md requires that the maintainer list, CODEOWNERS, and the repository's collaborator settings be updated together, so the documented roster cannot silently diverge from the enforced one.

Role changes: https://github.com/gma1k/podtrace/blob/main/GOVERNANCE.md#becoming-a-maintainer describes how maintainership is earned (a track record of merged non-trivial contributions, reviews that catch real problems, demonstrated compatibility judgment, and sustained engagement), with invitations extended by existing maintainers. Maintainers who step back are recorded as emeritus rather than removed.

Podtrace maintains a roadmap covering both intended work and explicit non-goals:
https://github.com/gma1k/podtrace/blob/main/ROADMAP.md

What the project intends to do: the roadmap lists ten items grouped into Now, Next, and Later horizons. Near-term work covers reconciling the documented CRD inventory, publishing a per-CRD v1beta1 readiness assessment, and completing the supply-chain baseline with SLSA provenance and an OpenSSF Security Insights manifest. Medium-term work covers defining a formal API deprecation policy, graduating individual CRDs from v1alpha1 to v1beta1, resolving whether to ship conversion webhooks, and making the importable Go API usable externally through runnable godoc examples. Longer-term work covers reaching v1.0.0 against the gates defined in https://github.com/gma1k/podtrace/blob/main/STABILITY.md, stating platform support in CI-verified versus best-effort tiers, and recruiting a second maintainer.

What the project intends not to do: the roadmap includes an explicit Non-goals section. Podtrace will not become a non-Linux agent, because its tracing core is eBPF and therefore Linux-kernel bound by construction. It will not become a policy engine, admission controller, or service mesh, because it observes workloads rather than admitting, mutating, or routing them. It will not implement agent self-update, so that agent version remains whatever the operator deploys and upgrades stay under the cluster owner's control. Non-goals are open to challenge through project Discussions.

Time horizon: the roadmap deliberately records themes and rough ordering rather than dates or quarters, on the stated reasoning that a roadmap with slipped quarters is worse than no roadmap. The scope covers well beyond one year: v1.0.0 is described as downstream of every other item and "not near," and the intervening work (per-CRD API graduation, a deprecation policy, a conversion story, supply-chain attestation, and tiered platform support) represents multi-year effort at the project's current capacity. Version-gating rules are kept separately in STABILITY.md so the roadmap states direction while STABILITY.md states commitments.

The roadmap is kept current: its status table reflects the shipped v0.14.x line and the 22 end-to-end chainsaw suites present in the repository. It is linked from the project README and cross-referenced from GOVERNANCE.md and CONTRIBUTING.md.

Podtrace documents its high-level design across a set of indexed architecture documents, entry point at https://github.com/gma1k/podtrace/blob/main/docs/README.md

Primary architecture document: https://github.com/gma1k/podtrace/blob/main/docs/architecture.md
Describes the system's major components and how they relate: eBPF programs running in the kernel (kprobes, uprobes, and tracepoints), a shared ring-buffer map carrying events across the kernel/userspace boundary, and a Go userspace layer comprising the tracer lifecycle manager, the object loader, the probe attachment layer, the event parser, and the cgroup filter that scopes events to a target pod. It documents the end-to-end data flow from kernel event through ring buffer, cgroup filtering, and enrichment, to the output surfaces (live CLI diagnostics, Prometheus metrics, and trace exporters).

Podtrace runs in three modes, each documented:

CLI: https://github.com/gma1k/podtrace/blob/main/docs/cli-architecture.md explains the node-spawn model. The CLI does not load eBPF on the invoking workstation; it identifies the node hosting each target pod and runs a short-lived privileged pod on that node, so the kernel loading eBPF is the same kernel running the target workload. The document states the structural reason: eBPF executes in the kernel where it is loaded, so a workstation kernel has no visibility into a remote node's processes or cgroups.

Operator and agent: https://github.com/gma1k/podtrace/blob/main/docs/operator.md documents the control plane. An operator Deployment runs three reconcilers, each owning one CRD. TracerConfigReconciler manages cluster-wide infrastructure including the agent ServiceAccount, RBAC, and the agent DaemonSet. PodTraceReconciler handles continuous tracing and aggregates per-node status. PodTraceSessionReconciler handles bounded diagnose sessions by spawning per-node Jobs. The agent DaemonSet runs one pod per node and performs the actual kernel-level tracing.

Per-session Jobs: documented in https://github.com/gma1k/podtrace/blob/main/docs/crd-podtracesession.md, including the per-session Role and RoleBinding the operator generates to scope report writes.

Kernel-side design detail is in https://github.com/gma1k/podtrace/blob/main/docs/ebpf-internals.md and the binary wire format crossing the kernel/userspace boundary is specified in https://github.com/gma1k/podtrace/blob/main/docs/event-schema.md

Key architectural properties are documented alongside the structure: the privilege model for each component (unprivileged operator, agent with CAP_BPF and CAP_PERFMON, session Jobs inheriting the agent's capability set) in https://github.com/gma1k/podtrace/blob/main/SECURITY.md; cgroup-based event isolation to the target pod; and cross-namespace consent gating enforced at both reconcile and admission, in https://github.com/gma1k/podtrace/blob/main/docs/cross-namespace-cr-targeting.md

Supported kernel, Kubernetes, and architecture matrices are in https://github.com/gma1k/podtrace/blob/main/docs/compatibility.md

Podtrace documents its security requirements, including both the guarantees it provides and the limits it explicitly does not cover, primarily in:
https://github.com/gma1k/podtrace/blob/main/SECURITY.md

What users can expect. The security model section states the privilege posture and blast radius of each component. The operator runs unprivileged (distroless nonroot, no host mounts); a compromised operator can only manipulate CRDs in its own namespace and cannot escalate to host privileges directly. The agent DaemonSet runs with CAP_BPF, CAP_PERFMON, and hostPID on each node, and the documentation states plainly that a compromised agent is roughly equivalent to a compromised privileged workload on that node. Per-session Jobs inherit the agent's capability set and are confined to the same single-node blast radius.

Tenant isolation is the central guarantee, because podtrace observes decrypted application-layer traffic. Cross-namespace tracing is consent-gated: a namespaced custom resource may target pods in a foreign namespace only when that namespace opts in via the podtrace.io/allow-tracing-from annotation. This is enforced by the operator at every reconcile, which is authoritative, and by a validating webhook at admission for fast feedback. Without the grant, a custom resource is confined to its own namespace and cannot capture another tenant's decrypted L7 traffic, DNS queries, syscalls, or file paths. Full semantics are documented at https://github.com/gma1k/podtrace/blob/main/docs/cross-namespace-cr-targeting.md

Supply-chain guarantees: all published artifacts (container image, Helm chart, release tarballs) are signed with cosign keyless via GitHub OIDC, with signing events recorded in the public Sigstore Rekor transparency log. Verification commands are documented at https://github.com/gma1k/podtrace/blob/main/docs/installation.md

What users cannot expect. SECURITY.md contains an explicit out-of-scope section. Issues in upstream Kubernetes, the host kernel's BPF verifier, and third-party exporters are not in scope. Attacks that presuppose an adversary already holding cluster-admin privileges are stated to be outside the threat model. Denial of service through excessive trace events on the agent's ring buffer is a user-side concern, with the bounding controls named (samplePercent and thresholds). Development-only paths are excluded.

Feature-level limits are documented the same way. https://github.com/gma1k/podtrace/blob/main/docs/crypto-detection.md states that AF_ALG crypto-socket detection is observation rather than enforcement: it surfaces use of the vulnerable interface, but does not block the syscall, does not detect exploitation, and is not a substitute for patching. It is also off by default.

Security-relevant defaults are documented at https://github.com/gma1k/podtrace/blob/main/docs/configuration.md, including that PII redaction (PODTRACE_REDACT_PII) and DNS-name stripping (PODTRACE_REDACT_DNS_NAMES) are disabled by default, and that plaintext export to non-loopback hosts is disabled by default and must be explicitly enabled.

Hardening guidance for deployments where attacker capability is a concern, covering image digest pinning, RBAC restriction of custom-resource creation, narrow granting of the cross-namespace annotation, and per-node session capacity caps, is in the hardening section of SECURITY.md, with per-distribution notes (SELinux, AppArmor, Pod Security Standards) in the platform guides under https://github.com/gma1k/podtrace/tree/main/docs

Supported versions and the backport window are stated at the top of SECURITY.md.

kubectl apply -f https://github.com/gma1k/podtrace/releases/latest/download/quickstart.yaml
kubectl apply -f https://github.com/gma1k/podtrace/releases/latest/download/quickstart-demo.yaml

podtrace -n production my-pod # realtime trace
podtrace -n production my-pod --diagnose 30s --export json # bounded diagnose with a JSON report

Prerequisites (kernel 5.8 or later, BTF availability, Kubernetes 1.28 or later) are stated in the installation guide and detailed at https://github.com/gma1k/podtrace/blob/main/docs/compatibility.md, with the full CLI reference at https://github.com/gma1k/podtrace/blob/main/docs/usage.md and a documentation index at https://github.com/gma1k/podtrace/blob/main/docs/README.md

The podtrace repository front page identifies and hyperlinks its achievements in a badge block at the top of the README, above the project description:
https://github.com/gma1k/podtrace#readme

The OpenSSF Best Practices badge is displayed and hyperlinked to the project's badge entry at https://www.bestpractices.dev/projects/12882

The badge is embedded using the dynamic badge image endpoint (https://www.bestpractices.dev/projects/12882/badge) rather than a static image. BadgeApp serves that endpoint with the project's current attainment level, so any change in badge level is reflected on the repository front page immediately and automatically, with no commit required. The 48-hour requirement is therefore satisfied structurally: the displayed badge cannot become stale relative to the recognition it represents.

Other achievements identified and hyperlinked in the same badge block:

OpenSSF Scorecard, linked to the project's live Scorecard report at https://scorecard.dev/viewer/?uri=github.com/gma1k/podtrace, also served as a dynamic badge reflecting the current score.
Artifact Hub, linked to the published Helm repository listing at https://artifacthub.io/packages/helm/podtrace/podtrace
eBPF Landscape, linked to the ebpf.io applications listing at https://ebpf.io/applications/
FOSSA license compliance and security status, linked to the project's FOSSA reports.
RepoGrade, linked to https://www.repo-grade.com/report/gma1k/podtrace
Go Reference (pkg.go.dev), release version, CI build status, test results, and test coverage.
Two further external acceptances are identified and hyperlinked in the README body, each of which required passing a third-party review process for inclusion: the OperatorHub.io community operator catalog (https://operatorhub.io/operator/podtrace) and the Kubernetes krew plugin index (https://krew.sigs.k8s.io/), through which podtrace is installable as kubectl krew install podtrace.

Podtrace produces a command-line tool and a Kubernetes operator. It has no graphical or web user interface of its own, so the criterion applies chiefly to terminal output and to project documentation.

Color is never the sole means of conveying information (WCAG 2.0 guideline 1.4.1). Podtrace's CLI emits no color at all: the codebase includes no terminal-color dependency and no ANSI color escape sequences in any output path. All diagnostic information is conveyed through text and document structure, so output remains fully intelligible when piped, redirected to a file, or read by a screen reader. Text contrast (guideline 1.4.3) is governed entirely by the user's own terminal configuration.

Text alternatives are provided for non-text content (guideline 1.1). Every image on the repository front page, including the project logo and all status badges, carries a descriptive alt attribute: https://github.com/gma1k/podtrace#readme. All documentation is authored in Markdown, which renders as semantic HTML and degrades cleanly to plain text.

All functionality is available from a keyboard (guideline 2.1). The CLI requires no pointer input, and the operator interface is declarative YAML applied with kubectl, so every capability is keyboard-driven and scriptable.

Terminal output integrity: podtrace displays data captured from traced application traffic, which is untrusted input. The project strips ANSI escape sequences from captured payloads before display, with test coverage for both color and cursor-movement injection. This prevents captured data from corrupting terminal state or disrupting assistive-technology output.

Screen-reader considerations: the optional real-time display refreshes by redrawing the report every five seconds, which produces overdraw. The project documents alternative output modes that avoid this entirely and provide identical information: bounded diagnose mode (--diagnose), structured JSON export (--export json), and the PodTraceSession custom resource, which writes its report to a Kubernetes ConfigMap or to object storage. These are documented at https://github.com/gma1k/podtrace/blob/main/docs/usage.md and https://github.com/gma1k/podtrace/blob/main/docs/viewing-events.md, so no capability is available only through the refreshing display.

The project ships an optional Grafana dashboard definition for its Prometheus metrics. Rendering and accessibility of that dashboard are governed by the user's own Grafana instance, which provides its own accessibility settings, rather than by podtrace.

Internationalization does not meaningfully apply to podtrace's output.

Podtrace is a kernel-level tracing tool. Its user-facing output consists overwhelmingly of technical identifiers that are fixed by the systems being observed and are not translatable: kernel symbol names (tcp_v4_connect, vfs_read, do_sys_openat2), syscall names, socket address families (AF_ALG, AF_INET), Linux capability names (CAP_BPF, CAP_PERFMON), protocol identifiers (TCP, UDP, DNS, HTTP/3, QUIC, gRPC), cgroup paths, filesystem paths, and Kubernetes API object kinds and field paths. These appear identically regardless of locale, because they are the names the kernel and the Kubernetes API themselves use.

The remaining prose is a thin layer of report section headings and summary labels wrapped around that technical content. Localizing it would degrade rather than improve usability: an operator reading a translated heading above untranslated kernel symbols must still search for those symbols in English in kernel documentation, source, and logs, so a partially translated report is harder to act on than a consistent one.

Locale-dependent formatting defects, which are the substantive risk this criterion guards against, are absent. All timestamps are emitted in locale-independent form: RFC 3339 in structured output and an explicit UTC pattern in report filenames. The codebase performs no locale-dependent number or date formatting. Sorting throughout the diagnostic layer is by event count, timestamp, or machine identifier rather than by locale-sensitive collation of human-readable text, so no incorrect ordering behavior exists in any locale.

Where podtrace's data is consumed by presentation layers that may serve a localized audience, it is emitted as structured data rather than prose: JSON export, OpenTelemetry traces (OTLP, Jaeger, Splunk, DataDog, Zipkin), and Prometheus metrics. Any downstream dashboard or integration can render that data in its users' language without podtrace being localized. See https://github.com/gma1k/podtrace/blob/main/docs/viewing-events.md and https://github.com/gma1k/podtrace/blob/main/docs/tracing-exporters.md

The project stores no passwords for authentication of external users.

Podtrace operates no website of its own. There is no documentation site, no GitHub Pages deployment, and no custom domain; documentation is Markdown rendered directly in the repository. All project sites are hosted by GitHub: the repository, issue tracker, discussions, and pull requests at https://github.com/gma1k/podtrace, release downloads at https://github.com/gma1k/podtrace/releases, and the container image and Helm OCI chart in GitHub Container Registry at ghcr.io/gma1k.

All inbound authentication for these surfaces is performed by GitHub, and this criterion states that the use of GitHub meets it. The project itself neither implements nor stores credentials for external user authentication.

Remaining project-related listings (Artifact Hub, OperatorHub.io, the krew index, pkg.go.dev, OpenSSF Scorecard, FOSSA) are third-party services not operated by the project.