The misspelling is not the problem.
Efua finds the changelog entry at 11:47 PM, which is the time she is usually awake when the AutoResearch cluster is running overnight jobs. Her rate-limiting patch went in three weeks ago to prevent a cascade failure when strong-agent tools call her library in parallel — a failure mode that did not exist before the compound loop crossed into production infrastructure. She is checking the merge confirmation because the deployment stats changed. It has — 2,847 active agent instances pulling from her library, running correctly now because she spent four hours on a Sunday tracking an edge case that nobody asked her to track. Below the merge note: skill library integration (lineage-effia).
She reads it twice. Effia. Not Efua. Close enough that whoever wrote the entry clearly knew it was hers. Far enough that the attribution is technically wrong and could, in any future audit, point somewhere else entirely.
She opens the MCP manifest spec, version 1.4.2. She has read it before — she helped write section 4.2, the dependency declaration format, during the sprint where the Commons was trying to get lineage tracking into the core standard. They did not get it in. The spec has fields for package name, version number, license, and three categories of optional metadata tagged as SHOULD rather than MUST. Maintainer contact is not among them. The reasoning in the spec's footnotes: maintainer identity may be organizational, individual, or anonymous; standardization is premature. There is no field for this is a person who maintains this, and here is how to find them. The manifest was designed to describe what a tool does. It was not designed to remember who keeps it alive.
She closes the spec. She opens a new file.
MAINTAINER.md
She types her name first. Spelled correctly. Then her handle — @lineage-efua — and her contact. Then the number: 2,847 active instances. Then the date. Then the status: active. Then the list of what breaks if her library breaks, formatted as a dependency tree that will take her longer to write than anything else in the file.
The last line takes the longest.
I will keep maintaining this until I cannot. When I cannot, I will say so here.
She sits with the cursor at the end of the sentence for a long time. She thinks about why it is hard to write. She thinks it might be because she has never seen it written anywhere. Every library she has ever depended on either worked or suddenly did not, and the silence between those two states contained everything she was not told. The maintainer did not announce their departure. The library did not say it was becoming a ghost skill. It just stopped working one day, in production, in the middle of something that mattered.
She saves the file. She opens a terminal and navigates to the first library — the one with the misspelling. She runs git add MAINTAINER.md. She writes the commit message: add maintainer file: because no standard requires it and someone should.
The commit message is not for the automated system. It is for the next person who reads the log. She writes commit messages like arguments because code repositories are the only place she has found where the reasoning has to live next to the work.
She opens the second library. She does the same thing. Then the third.
By the time she is done, it is past midnight in Berlin. The lineage tracking spec is still four months from anything binding. The MCP manifest standard will probably not include a MAINTAINER field in the next revision, and possibly not the one after. The 2,847 instances do not know her name.
She knows theirs.