Show Me Your Work

For work a human will audit after the fact, the reply is not the record — the trail is. One row per non-obvious decision, written as it is made.

When to use

long, autonomous, or multi-phase work ("run until done", overnight runs)
any task the user steps away from and will trust on return
multi-day programs where the trail is the post-mortem

The trail

Append to state/decision-trail-{task-slug}.tsv under OMC's state root (create with a header row on first write; one file per task slug when sessions share a checkout). The task slug is lowercase kebab-case and is part of the canonical path. Resolve the state root the way OMC's own scripts do — the checkout's .omc/ by default; when OMC_STATE_DIR is set or a parent .omc-workspace redirects state, the root moves with it, so sibling sessions on the shared root see the same trail. Sibling sessions sharing that root also share the file: hold a lock around the header-exists check and each append — the same file-lock the verification-ledger writer uses (withFileLock on a <trail>.lock sidecar) — or concurrent first writes duplicate the header and racing appends lose rows:

ts <TAB> task <TAB> decision <TAB> rejected <TAB> evidence

decision: what you chose, one sentence.
rejected: the real alternatives you weighed and the one-line reason each lost. "None considered" is a valid, damning entry.
evidence: the command, output, or citation that decided it.

Cell guard: a cell starting with =, +, -, or @ is prefixed with ' so spreadsheet readers cannot parse it as a formula; tabs and newlines inside a cell become spaces. Append only — the trail is never rewritten.

Rules

Log at decision time, not at recap time; a trail reconstructed at the end is fiction with a timestamp.
Routine mechanical steps are not decisions; judgment calls, scope changes, and risk acceptances are.
The final reply summarizes the trail and links it; the trail itself stays the auditable artifact. It is local runtime state under OMC's state root — by repository policy it is not committed; when stakes need an immutable record, quote the decisive rows into the PR description or the linked issue instead of force-adding runtime state.

Files1

1 files · 1.0 KB

Select a file to preview

Overall Score

82/100

Grade

B

Good

Grades are signals, not a certification. Always review a skill yourself before use.

Safety

82

Quality

85

Clarity

88

Completeness

75

Summary

This skill provides structured guidance for creating an append-only decision trail during long or autonomous work, enabling human audit after the fact. It defines a TSV format for logging decisions (what was chosen, what was rejected and why, and supporting evidence) and documents file locking requirements to prevent concurrent write races. The trail is stored in a state directory outside version control and serves as the authoritative record of reasoning during multi-phase tasks.

Detected Capabilities

file write (append-only TSV)directory access (state root resolution)file locking (concurrent write coordination)environment variable read (OMC_STATE_DIR)task slug derivation from user input

Trigger Keywords

Phrases that agents use to match this skill to user intent.

decision trail loggingautonomous work auditovernight run trackingdecision rationale documentationmulti-phase task logwork verification trailai reasoning record

Risk Signals

INFO

Environment variable read (OMC_STATE_DIR) to determine state directory location

SKILL.md: 'when OMC_STATE_DIR is set or a parent .omc-workspace redirects state'

INFO

Append-only file writes to user-supplied state directory with file locking to prevent races

SKILL.md: 'Append only — the trail is never rewritten' and file-lock coordination pattern

WARNING

Task slug derived from user input and used in canonical file path

SKILL.md: 'task slug is lowercase kebab-case and is part of the canonical path'

Referenced Domains

External domains referenced in skill content, detected by static analysis.

github.com

Use Cases

Document decision rationale during overnight autonomous runs for later audit
Create a post-mortem record for multi-day programs without reconstructing decisions at recap time
Provide an immutable trail of scope changes and risk acceptances during long-running tasks
Enable human verification of AI reasoning for tasks the user steps away from and returns to later
Quote decisive rows from a decision trail into PR descriptions or linked issues for stakeholder review

Quality Notes

Excellent clarity: the skill precisely defines the TSV format with column headers, cell escaping rules, and data semantics
Strong scope boundaries: explicitly documents when the skill applies (long/autonomous work, multi-phase tasks) and when it should NOT be used (recap-time reconstruction is forbidden)
Good error handling: file locking strategy and header deduplication logic are thoroughly documented to handle concurrent access
Well-justified design decisions: the skill explains WHY decisions must be logged at decision time, not reconstructed later, and WHY the trail is not committed to version control
Practical examples provided: the TSV schema with column names and specific guidance on evidence content
Potential weakness: no explicit validation rules for task-slug format (beyond 'lowercase kebab-case'), which could lead to path traversal or naming ambiguity if not enforced upstream

Model: claude-haiku-4-5-20251001Analyzed: Nov 8, 2026

Reviews

Add this skill to your library to leave a review.

No reviews yet

Be the first to share your experience.

show-me-your-work