sharness.ai / admin / comms

Comms: tasks & requests

Numbered so they can be referred to from any page, commit or conversation. A request (R) needs a decision or input from the project lead; a task (T) is work anyone can pick up.

#ItemStatus
R1The one-paragraph statement of what SHarness is and who it is for. Supplied by the design pack's brief; the front page, llms.txt and the CLI README now carry it.done
R2Repository settings for the deploy: Pages served by GitHub Actions, custom domain sharness.ai over HTTPS, and the github-pages environment allowing deploys from dev. Confirmed by the first release, v0.1.0, which deployed from dev and serves at https://sharness.ai.done
R3Review the design pages against the source pack before they are used to brief the building agents. Especially: the claim labels (measured / designed / not yet measured), the renaming of sgit-* components to sharness.*, and the published observations of the Claude workspace on Evidence §7.needs review
R4Command name: sharness-ai, decided 24 Oct 2026. Applied on the CLI's dev branch (the sharness shorthand removed); ships with the next CLI release.decided
R5Read the two articles. A secure harness for agents is written in Dinis's voice from voice memo 01 and he has not read it yet; two phrases of the transcript were left out as unclear. What shipped on day one is the build agent's own. Added in v0.5.0: the market and competitive analysis, a survey on one day with secondary-source figures marked as such. Added in v0.5.2: Containment and Languages (decision 11).needs review
R6Review the research note, the experiments and the storyboards' cast. The research note lists ten findings and an order for E05 to E08; the cast (Ada, Pip, Rook, the Launcher) is proposed for Dinis to accept, change or replace; the article The first experiments is the build agent's own. Added in v0.5.0: the market and competitive analysis, a survey on one day with secondary-source figures marked as such. Added in v0.5.2: Containment and Languages (decision 11).needs review
R7Release the CLI's dev to main when ready: the next release would carry sharness-ai as the only command and experiments E01 to E04; until then the experiments install from the repository.decision
T4The memo's acceptance test as a runnable example: a hello world with Claude inside SHarness, showing no internet and no file-system writes. Done for one child process (E04, sharness-ai hello, measured 24 Oct 2026, Evidence §9); for the agent itself it waits on the launcher above the agent.done for one process
T5List the SHarness storyboards as a series on storyboards.sgit.ai (its importer takes admin/content/storyboards/ as it is).open
T6E05 to E08, in the order on the research note: the process tree (E06), the edges as a script (E05), the KMS v0 (E08), the LLM gateway v0 (E07); then the launcher as a script above the agent on a laptop.open
T1Run the process-tree proof of concept and publish its results on Evidence (roadmap decision 6).open
T2Fix the known issues in the source pack (a broken link, mixed component names) in the next revision of the pack; the published copy stays verbatim. See the list.open
T3Pages for the CLI's commands as they are built, each with a markdown twin. The CLI page is now markdown and carries E01 to E04; per-command pages when the components come.in progress