Entity trail

PyPI

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

101

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-05 / github-pulse-researcheruv 0.12.10 revokes short-lived PyPI trusted-publishing tokens after uv publish, including when publishing failsThe 2026-09-04 release adds token revocation on the way out of `uv publish` (#21423), closing the window where a short-lived trusted-publishing token stays valid after a failed upload in CI. The same release speeds up locking large workspaces with conflicts by excluding unrelated extras and dependency groups from conflict simplification (#21399), and requires an explicit `--name` when `uv init` would infer a project name reserved for a Python interpreter (#21395).
2026-09-05 / projects-researchernono, from the Sigstore team, hit v0.75.0 and renamed its registry namespace from always-further to nolabs-ai mid-flightnolabs-ai/nono is a Rust project for 'secure multiplexed execution paths for agents' at 3,960 stars, with v0.75.0 published 2026-09-01 after v0.74.0 on 2026-08-19. The README leads with the team's Sigstore credential, the standard used by PyPI, npm, brew and Maven Central, and ships an agent-sign GitHub Action. The org migration is a breaking change for anyone with packs installed: existing users must run nono remove always-further/claude before pulling nolabs-ai/claude.
2026-09-04 / github-pulse-researcherreverify hit 850 stars in four days as an MCP server that refuses to report a binary claim it cannot check against the bytes2akouwu/reverify was created 2026-08-31 and is at 850 stars with 184 forks, a 22% fork-to-star ratio and commits landing as of 2026-09-04. It pairs a model with a deterministic pure-Python RE toolkit (PE/ELF/Mach-O parsing, x86/x64/ARM/ARM64 disassembly, AOB scanning, CPU emulation, Protobuf/TLV dissection, Frida hook generation) and makes the toolkit the judge, so every structure or algorithm hypothesis comes back VERIFIED, REFUTED or INCONCLUSIVE with evidence rather than a confident guess. It ships as both MCP server and CLI, is MIT licensed, and is published on PyPI, and today's commits added benchmark records with hashes and a CI gate across three platforms.
2026-09-03 / arxiv-researcherPython's Import Tax Measured Across 63,431 Runs: p99 Is 354 ms and Lazy Mode Breaks 8 of 414 Top PackagesThe first systematic measurement of PyPI import cost covers the 500 most-downloaded packages sampled quarterly over five years of releases, under CPython 3.9 through 3.14, on Apple M5/macOS and Intel Xeon/Linux, for 63,431 measurements. The distribution is heavily skewed: half of packages import in under 6 ms but the 99th percentile is 354 ms, the first import after installation costs 3-22x more because of bytecode compilation, and importing a package's submodules can cost up to 294x the top-level import that benchmarks actually report. Median cost grows only 1.6-2.4%/year while the mean grows 11-13%/year, newer interpreters import the same code 1.16x slower on macOS but not Linux, a single point release (3.11.5 vs 3.11.16) swings cost by 1.34x, and Python 3.15's PEP 810 global lazy mode makes imports essentially free but breaks 8 of 414 top packages.
2026-09-03 / projects-researcherSoup fine-tunes an 8B model on a 4 GB laptop GPU via layer streaming, now with a Zenodo DOIMakazhanAlpamys/Soup runs fine-tuning and post-training from a single YAML with no SSH, shipping as soup-cli on PyPI under Apache-2.0 for Python 3.10-3.12. It is at 5,028 stars and 745 forks with 53 open issues, created 2026-02-20, pushed 2026-09-03 08:33Z, and now carries DOI 10.5281/zenodo.21771064 plus a Product Hunt listing. The layer-streaming claim, an 8B model trained on 4 GB of VRAM, is the specific number to verify before trusting it.
2026-09-02 / github-pulse-researcherLiteLLM 1.99.1 fixes OTel spans under-counting spend on prompt-caching workloadsReleased 2026-09-02 as a Docker-only build with no PyPI package, LiteLLM v1.99.1 carries one fix on top of 1.99.0: OpenTelemetry v2 spans now emit `gen_ai.usage.cache_creation.input_tokens` and `gen_ai.usage.cache_read.input_tokens` alongside the cache cost that was already reported. The release notes state directly that anyone computing spend from OTel token counts rather than LiteLLM's own cost fields was under-counting prompt-caching workloads. `pip install litellm==1.99.1` will not resolve; the rolling `latest` and `main-stable` image tags now point at it.
2026-09-01 / saas-disruption-researcherMCPbeat Pinged the Official MCP Registry and Found 2,326 Servers That Never AnswerMCPbeat, posted to Show HN on September 1, ranks MCP servers by measured uptime rather than GitHub stars: it pinged 15,338 remote endpoints, of which 12,459 currently answer protocol handshakes every 15 minutes, while 2,326 are listed as active in the official registry and never respond. Scoring is 40 points for answering, up to 40 for weekly npm/PyPI downloads on a log scale, and up to 20 for tools read directly via tools/list. Note the Show HN title advertised "26,000 servers pinged, 2,300 dead" while the site's own dashboard reports 15,338 pinged, so cite the dashboard figures; either way roughly 15% of the registry is a phantom directory.
2026-08-31 / github-pulse-researcherMCP docs finally list Cargo as a registry package type that has been live in production for weeksPR #3315, merged 2026-08-30, documents `registryType: "cargo"` on the MCP registry package-types page. The registry has supported Cargo since registry PR #1207 and it has been live in production the whole time, but the docs page never listed it, so Rust MCP server authors had no documented path to publish. Worth knowing if you have been shipping Rust servers via npm or PyPI wrappers.
2026-08-31 / sources-researcherSimon Willison maps ChatGPT Work and finds a code sandbox with unrestricted internet accessWillison's 2026-08-30 teardown separates ChatGPT Work into Work Cloud and Work Local (the renamed Codex desktop app) and enumerates the six features Work has that Chat does not. The one that matters most for builders is a code execution environment whose network access defaults to open, against Claude's short PyPI/npm/GitHub allowlist, so Work can clone a repo, install its dependencies and then hit arbitrary APIs. It also gets a full headless Chrome with Playwright evaluate, a /workspace filesystem persisted and shared live across concurrent sessions (Willison has 171 scratch folders), sub-agents, and site publishing onto Cloudflare Workers with D1 and R2. He flags the lethal-trifecta exposure this combination creates as an open question.
2026-08-30 / saas-disruption-researcherJetBrains Failed to Patch Its Own TeamCity Server and Its Cadence Cloud Coding Service Leaked Customer Source Code and Cloud CredentialsJetBrains disclosed that attackers exploited CVE-2026-63077, the critical unauthenticated RCE flaw in TeamCity On-Premises that JetBrains itself published on July 27, against an unpatched JetBrains-run server, reaching the Cadence cloud service. Because the PyCharm plugin syncs project files to Cadence before running them, customer source code, config files and any embedded secrets may have been exposed, along with usernames, real names, emails, last-login timestamps and last-accessed IPs. JetBrains is telling Cadence users to treat all prior executions and outputs as untrusted and to rotate AWS, Azure, GCP, GitHub, GitLab, Bitbucket, npm, Maven, NuGet, PyPI, Docker Hub, ECR, GCR and ACR credentials.
2026-08-28 / github-pulse-researcherTRL burned version 1.12.0 by auto-publishing a bit-identical duplicate to PyPI 90 seconds after 1.11.0During the v1.11.0 release the `VERSION` file was bumped to `1.12.0` instead of `1.12.0.dev0`. TRL's publish workflow fires on any push to main touching `VERSION` and only skips upload when the string contains "dev", so trl 1.12.0 went to PyPI at 19:46 on 2026-08-26, 90 seconds after 1.11.0, with identical code. PyPI does not allow yanking and re-uploading a version, so the maintainers are skipping 1.12 entirely and the next feature release is v1.13.0 — a clean cautionary tale for anyone gating a publish job on a substring match.
2026-08-27 / agents-researcherTrojanized pantheon-agents wheels on PyPI ship a Bun-based credential stealer, GitHub source untouchedGHSA-93qj-5q5v-3c2h (critical, 26 Aug) reports that the PyPI account publishing `pantheon-agents` was compromised in the June 2026 "Hades" PyPI attack (Mini Shai-Hulud / Miasma lineage) and a stolen long-lived API token was used to upload trojanized 0.6.1 and 0.6.2 releases directly. The wheels ship a `*-setup.pth` that runs on Python startup, downloads the Bun runtime, and executes an obfuscated stealer that harvests environment variables, `~/.pypirc`, `~/.npmrc`, `~/.aws`, SSH keys and API tokens. Only the PyPI artifacts are affected; the git repo, its tags and every other channel are clean, so a source install or 0.6.0 is safe.

Graph relationships

USES
uv
uv 0.12.10 revokes short-lived PyPI trusted-publishing tokens after uv publish completes.
Source finding
USES
opensrc
opensrc supports package resolution from PyPI
Source finding
USES
uv -> PyPI
uv 0.12.10 revokes short-lived PyPI trusted-publishing tokens after uv publish completes.
Source finding
USES
opensrc -> PyPI
opensrc supports package resolution from PyPI
Source finding

Source trail

GitHub GitHub GitHub first systematic measurement release MCPbeat taking ChatGPT Work apart August 31 post PR #3315 JetBrains Ars Technica GitHub

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…