SShortSingh.
Back to feed

Fixing AI agents and git pushing to the wrong GitHub account with per-folder config

0
·9 views

Developers using several GitHub accounts on one machine often find that commits, pushes, and pull requests from tools like Claude Code go out under the wrong identity. The article explains that three separate settings decide the account: the git commit author, the credential used for pushes, and the GitHub CLI's active account. Common workarounds such as SSH host aliases, conditional email config, switching accounts manually, or using direnv with tokens each cover only part of the problem. The proposed fix organizes projects into per-account folders and uses git's includeIf directive along with a custom credential helper that pulls the right token from the GitHub CLI. The author also notes that git can select SSH keys per repository, which removes the need for host aliases when using SSH remotes.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Smartlook Shuts Down in 2027, Prompting Users to Plan Data Exports and Alternatives

Smartlook stopped selling new subscriptions on 31 May 2026 and will switch off its platform on 30 September 2027, deleting all remaining customer data. Existing paid plans end earlier, so users should note their own expiry date and plan a switch before it. The company, now part of Cisco, is folding its features into Digital Experience Analytics within Splunk Observability Cloud, and Splunk will not migrate historical data. Smartlook advises exporting retained data via its APIs before the paid subscription expires and no later than 31 August 2027. The source recommends listing key events, funnels, reports and recordings to keep, then trialing replacement tools, including self-hosted WordPress options, before the deadline.

0
ProgrammingDEV Community ·

Experienced engineers face late-stage hiring rejections as 2026 tech hiring tightens

A DEV Community analysis argues that the 2026 engineering hiring slowdown reflects structural employer incentives rather than a lack of skill among experienced engineers. It contends that employers often seek Staff-level scope while offering Senior-level compensation, which makes late-stage rejections a predictable result of current hiring practices. The author notes that multi-round interview funnels compound filter risk, so strong candidates can be screened out even when their capability is sound. Late-stage veto decisions, driven by conservative risk-averse rubrics and noisy evaluation signals, are described as frequent causes of final-round rejections. The piece recommends that senior candidates shift from high-volume applications toward more targeted, leverage-focused job searches.

0
ProgrammingDEV Community ·

Crypto Swap Interfaces Need More Than a Confirm Button, Developer Argues

A developer article argues that a crypto swap involves much more than a simple select-and-confirm flow, spanning wallet connectivity, transaction preparation, token permissions, network resources, and blockchain confirmation. It notes that a connected wallet does not guarantee a transaction is ready, since network, balance, token support, and resource checks still matter. The piece stresses that token approvals are a separate authorization step that users should clearly understand before signing. It also explains that swap quotes are estimates that can shift due to liquidity, price movement, and network conditions, making slippage tolerance important. The article is framed around design lessons for Web3 applications, particularly those using TRON's TRC-20 tokens.

0
ProgrammingDEV Community ·

Apify MCP tool definitions and dataset results add significant token overhead

A developer enabled ten Apify Actors on the hosted MCP server and counted 27,797 tokens in their tool definitions, with the full array including shared tools reaching 31,598. Fetching three rows from each Actor's default dataset produced results ranging from about 1,231 to 7,434 tokens. Requesting only the needed fields cut one rental-listing result from 7,434 to 298 tokens, a 96% reduction, while preserving null values. The author suggests using workflow-specific connections to limit tool lists and projecting fields in dataset requests to reduce payload size. These are payload measurements using a common tokenizer, and actual model costs depend on how each client loads tools and handles MCP results.