# Tekton v1.3.3

2026-03-18 · [Original release notes](https://github.com/tektoncd/pipeline/releases/tag/v1.3.3) · [Full summary](https://ratatosk.io/en/releases/tekton/v1.3.3)

Security 2 (critical 1) · Breaking changes 1 · Important fixes 3 · Enhancements 1 · Other changes 4

## Security (2)

### [critical] Git resolver path traversal fixed

- Affects: setups using the git resolver
- [CVE-2026-33211](https://www.cve.org/CVERecord?id=CVE-2026-33211), [GHSA-j5q5-j9gm-2w5c](https://github.com/tektoncd/pipeline/security/advisories/GHSA-j5q5-j9gm-2w5c)

A user able to create a `ResolutionRequest` using the git resolver could previously read arbitrary files from the resolver pod, including ServiceAccount tokens. Requests with traversal components in `pathInRepo` or symlink-based escapes are now rejected.

> - **[GHSA-j5q5-j9gm-2w5c](https://github.com/tektoncd/pipeline/security/advisories/GHSA-j5q5-j9gm-2w5c)** (Critical): Path traversal in git resolver allows reading arbitrary files from the resolver pod. Fixed by validating the `pathInRepo` parameter to prevent directory traversal.

### [medium] Controller panic from long resolver names fixed

- Affects: TaskRuns or PipelineRuns using custom resolver names of 31 characters or more
- [CVE-2026-33022](https://www.cve.org/CVERecord?id=CVE-2026-33022), [GHSA-cv4x-93xx-wgfj](https://github.com/tektoncd/pipeline/security/advisories/GHSA-cv4x-93xx-wgfj)

Before this release, such names could panic the controller into a restart loop and block cluster-wide reconciliation. Now the resolver-name prefix is truncated to fit the DNS-1123 label limit while the full hash is preserved.

> - **[GHSA-cv4x-93xx-wgfj](https://github.com/tektoncd/pipeline/security/advisories/GHSA-cv4x-93xx-wgfj)** / CVE-2026-33022 (Medium): Controller panic via long resolver name in TaskRun/PipelineRun. A user with permission to create TaskRuns or PipelineRuns could crash the controller into a restart loop by setting a resolver name of 31+ characters, causing denial of service cluster-wide. Thanks to @1seal for reporting this vulnerability.

## Breaking changes (1)

### PipelineRun child references sorted in status

- Affects: PipelineRuns with child references in status

Previously, an unsorted `childRef` array could trigger massive redundant status writes. Sorting the references mitigates those writes and the associated resource load and apiserver stability issues.

> * :bug: fix(pipelinerun): fix the issue of massive invalid status updates caused by unordered arrays, which will greatly impact the resource load and stability of the apiserver (#9312)

## Important fixes (3)

### `v1beta1` Pipeline validation panic fixed

- Affects: Pipelines using the `v1beta1` API with matrix parameters

Invalid result references in matrix parameters could cause a panic during validation. The webhook now returns a validation error for invalid expressions instead.

> * :bug: fix: panic in v1beta1 matrix validation for invalid result refs (#9260) — Resolved an issue where Pipelines with invalid result references in matrix parameters would cause a panic during validation (v1beta1 API)

### Git resolver token authentication fixed

- Affects: git resolver setups using `gitToken` and `gitTokenKey` for HTTP token authentication

When resolving a private repository, cloning failed with an `authentication required` error because the resolver did not authenticate with the provided token. Git operations now use `gitToken` for remote authentication.

> * :bug: fix(#8940): Fix token-authentication header in git resolver (#9115) — Before this change, there was a regression in which the git resolver was not authenticating with the provided `gitToken` and `gitTokenKey`, breaking the git resolver's http token-based auth. After this change, all git operations performed by the git resolver use the provided `gitToken` for remote authentication.

### Immediate requeue loops for runs with disabled timeouts fixed

- Affects: PipelineRuns and TaskRuns with timeouts disabled, including setups with `default-timeout-minutes` set to "0"

Previously, these runs could trigger excessive reconciliation instead of waiting for watch events. The reconciler now waits for watch events when no timeout is configured, while PipelineRuns still requeue to enforce configured pipeline, task, or finally timeouts.

> * :bug: fix: Prevent excessive reconciliation when timeout disabled (#9355)

## Enhancements (1)

- `hostUsers` support in TaskRun and PipelineRun `PodTemplate`

## Other changes (4)

- Added `ci-summary` fan-in job to the `release-v1.3.x` CI workflow
- Release pipeline plumbing revision updated
- Full image references in the release pipeline `precheck` task
- CI test reliability improvements backported

---

Summaries are produced by AI, so they can be wrong, and they are provided without warranty.
