What happened
After an owner correction, the coordinating session designed a PreToolUse hook (a gate that blocks code edits unless a decision-trail row exists) inline, without routing through how. The gate shipped with the wrong shape: it keyed on the session instead of the unit, and it guarded the one writer that is forbidden to write code while every sanctioned writer (delegates in worktrees, the external runner) ran outside its reach. A reflect pass flagged that how's description lists placement and ownership questions ("where should this live", "which package owns this", "is this the right layer") but nothing that reads as "I am about to build a mechanism for the harness itself", so the skill never triggered.
Proposal
Add to the how description's trigger list (and the Placement bullet in poteto-mode/SKILL.md if you keep the two in sync): designing an enforcement mechanism, hook, gate, or lint for the agent harness itself, where the questions are which actor it guards, what it keys on, and which writers pass outside it. Optionally a one-paragraph note in Critique mode that harness mechanisms get the same rubric as product code: what does it hide, who does it guard, what escapes it.
Why upstream
A description tune is a skill-body change; this project does not fork pstack and keeps its own rules outside the plugin. Context: open-pstack 1.2.1 on Claude Code, macOS.
What happened
After an owner correction, the coordinating session designed a PreToolUse hook (a gate that blocks code edits unless a decision-trail row exists) inline, without routing through
how. The gate shipped with the wrong shape: it keyed on the session instead of the unit, and it guarded the one writer that is forbidden to write code while every sanctioned writer (delegates in worktrees, the external runner) ran outside its reach. Areflectpass flagged thathow's description lists placement and ownership questions ("where should this live", "which package owns this", "is this the right layer") but nothing that reads as "I am about to build a mechanism for the harness itself", so the skill never triggered.Proposal
Add to the
howdescription's trigger list (and the Placement bullet inpoteto-mode/SKILL.mdif you keep the two in sync): designing an enforcement mechanism, hook, gate, or lint for the agent harness itself, where the questions are which actor it guards, what it keys on, and which writers pass outside it. Optionally a one-paragraph note in Critique mode that harness mechanisms get the same rubric as product code: what does it hide, who does it guard, what escapes it.Why upstream
A description tune is a skill-body change; this project does not fork pstack and keeps its own rules outside the plugin. Context: open-pstack 1.2.1 on Claude Code, macOS.