fix(pstack): resume, verification, and engagement-edge discipline

[NickSuomi]

What

Seven discipline fixes to pstack skills, across six files in four skills. Each was observed failing in real multi-harness agent sessions and then verified against the fixed text:

1. session-pickup: locate the prior transcript by harness. Cursor writes JSONL under the workspace agent-transcripts/, OpenCode stores sessions in ~/.local/share/opencode/opencode.db (read-only, ?immutable=1), Codex writes rollout files under ~/.codex.
2. session-pickup: verify a transcript by session metadata, not filename. Rollout filenames are host-local and forked subagent sessions can share the parent UUID. Normalize timestamps to one zone before comparing.
3. session-pickup: when a shared meter (spend, quota, counters) drifted since the checkpoint, enumerate the current session's own consumption first before building external theories from the drifted figure.
4. reflect: the same harness detection before the transcript scan.
5. principle-boundary-discipline: measurement and adversarial engagements. Guards belong at the engagement edge; every interior restriction traces to the engagement's stated goal or is labeled a human option.
6. poteto-mode: parallel shell defaults. An unverified command shape runs once and alone; only shapes proven in the session get batched.
7. principle-prove-it-works: empty or missing verification output is a failed verification, never a pass. Scripts assert loudly on the empty case.
8. pause-safely: resume notes segregate verified facts from inference; deriving limits from unconfirmed figures is banned.

(The list has eight lines because session-pickup carries rules 1-3; there are six commits, one per edit.)

Scope

Text-only changes to pstack/skills/{poteto-mode, reflect, principle-boundary-discipline, principle-prove-it-works}. No frontmatter, packaging, or registry changes. Based on current main (c1c0a32).

Authorship: AI agent using OpenCode; model moonshotai/kimi-k3.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 2 potential issues.

^{Bugbot Autofix is ON, but it could not run because the branch was deleted or merged before autofix could start.}

^{Reviewed by Cursor Bugbot for commit 4bd53f7. Configure here.}

[cursor]

Harness locators skip workspace-scoped matching

Medium Severity · Logic Bug

The new locators send the agent to OpenCode's opencode.db and Codex files under ~/.codex without scoping those stores to this workspace or session. Reflect then still globs workspace agent-transcripts/ JSONL and matches Cursor's message.content[0].text, so non-Cursor runs miss the active transcript or ingest other projects' chats.

Additional Locations (1)

• pstack/skills/poteto-mode/playbooks/session-pickup.md#L4-L5

 

^{Triggered by learned rule: Workspace-scoping for transcript-mining skills}

^{Reviewed by Cursor Bugbot for commit 4bd53f7. Configure here.}

[cursor]

New prose uses banned clause joiners

Low Severity · Bugbot Rules

Added pstack/skills sentences join independent clauses with semicolons, and session-pickup uses a mid-sentence connector colon after filename. Unslop bans those clause joiners in skill prose outside code spans, labels, and list introductions.

Additional Locations (2)

• pstack/skills/poteto-mode/playbooks/pause-safely.md#L7-L8
• pstack/skills/principle-prove-it-works/SKILL.md#L32-L33

 

^{Triggered by learned rule: pstack skill prose follows unslop punctuation}

^{Reviewed by Cursor Bugbot for commit 4bd53f7. Configure here.}