Files
boshu2__agentops/docs/task-proposals-2026-04-29.md
T

2.5 KiB

Task Proposals (2026-04-29)

The following tasks are based on direct file inspection in this repository and are intentionally scoped for execution in one PR each.

1) Typo fix task

Title: Fix recieve misspelling in quick-fix workflow example

Evidence: docs/workflows/quick-fix.md includes recieve in both the user quote and example commit message.

Scope:

  • Change recieve → receive in the two example lines.
  • Keep the rest of the example unchanged.

Acceptance criteria:

  • rg -n "recieve" docs/workflows/quick-fix.md returns no matches.
  • Example remains a typo-fix walkthrough.

2) Bug fix task

Title: Avoid non-digit phase labels in daemon reconcile tests when phase >= 10

Evidence: Test IDs are built using ASCII arithmetic (string(rune('0'+phase))) in cli/internal/daemon/reconcile_test.go. For two-digit values this generates punctuation (: etc.) instead of numeric strings.

Scope:

  • Replace rune math with integer formatting (strconv.Itoa(phase) or fmt.Sprintf("%d", phase)).
  • Update all affected test-field builders in that file.

Acceptance criteria:

  • IDs/labels in tests render correctly for phase values above 9.
  • cd cli && go test ./internal/daemon -run Reconcile passes.

3) Documentation discrepancy task

Title: Remove duplicate install command in AGENTS install section

Evidence: AGENTS.md repeats the same install.sh command twice under “Other agents (for example Cursor): install only selected skills”.

Scope:

  • Deduplicate repeated command.
  • Clarify whether the command installs all skills or selected skills, matching actual behavior.

Acceptance criteria:

  • No duplicate command remains in the subsection.
  • Wording aligns with command behavior.

4) Test improvement task

Title: Add regression test for tool-event scanners to ignore plain-text mentions

Evidence: cli/testdata/transcripts/real-2.4mb.jsonl contains both regular content strings and structured tool events; scanners that grep raw text can false-positive on tool-name mentions.

Scope:

  • Add/extend a parser-level test that distinguishes:
    • structured tool events ("type":"tool_use" + event shape), and
    • incidental mentions inside free-form text.
  • Use existing transcript fixtures or add a compact targeted fixture.

Acceptance criteria:

  • Test fails with naive string matching and passes with structured-event parsing.
  • Existing behavior for real tool-use events remains covered.